Merge 399010fd12 into d700b56c9c

This reverts commit 6fe0a8784a.

Revert the commit now that the issue is fixed.

Signed-off-by: CHEN, CHUN <jim60105@gmail.com>

.github/workflows/build-and-release.yml

@@ -0,0 +1,34 @@
+name: Build and release
+
+on:
+  release:
+    types: [published]
+
+jobs:
+  build:
+    runs-on: ubuntu-latest
+    steps:
+      - name: Checkout
+        uses: actions/checkout@v4
+
+      - name: Install uv
+        uses: astral-sh/setup-uv@v5
+        with:
+          version: "0.5.14"
+          python-version: "3.9"
+
+      - name: Check if lockfile is up to date
+        run: uv lock --check
+
+      - name: Build package
+        run: uv build
+
+      - name: Release to Github
+        uses: softprops/action-gh-release@v2
+        with:
+          files: dist/*.whl
+
+      - name: Publish package to PyPi
+        run: uv publish
+        env:
+          UV_PUBLISH_TOKEN: ${{ secrets.PYPI_API_TOKEN }}

.github/workflows/python-compatibility.yml

@@ -0,0 +1,34 @@
+name: Python Compatibility Test
+
+on:
+  push:
+    branches: [main]
+  pull_request:
+    branches: [main]
+  workflow_dispatch: # Allows manual triggering from GitHub UI
+
+jobs:
+  test:
+    runs-on: ubuntu-latest
+    strategy:
+      matrix:
+        python-version: ["3.9", "3.10", "3.11", "3.12"]
+
+    steps:
+      - uses: actions/checkout@v4
+
+      - name: Install uv
+        uses: astral-sh/setup-uv@v5
+        with:
+          version: "0.5.14"
+          python-version: ${{ matrix.python-version }}
+
+      - name: Check if lockfile is up to date
+        run: uv lock --check
+
+      - name: Install the project
+        run: uv sync --all-extras
+
+      - name: Test import
+        run: |
+          uv run python -c "import whisperx; print('Successfully imported whisperx')"

.gitignore

@@ -1,2 +1,171 @@
-whisperx.egg-info/
-**/__pycache__/
+# Byte-compiled / optimized / DLL files
+__pycache__/
+*.py[cod]
+*$py.class
+
+# C extensions
+*.so
+
+# Distribution / packaging
+.Python
+build/
+develop-eggs/
+dist/
+downloads/
+eggs/
+.eggs/
+lib/
+lib64/
+parts/
+sdist/
+var/
+wheels/
+share/python-wheels/
+*.egg-info/
+.installed.cfg
+*.egg
+MANIFEST
+
+# PyInstaller
+#  Usually these files are written by a python script from a template
+#  before PyInstaller builds the exe, so as to inject date/other infos into it.
+*.manifest
+*.spec
+
+# Installer logs
+pip-log.txt
+pip-delete-this-directory.txt
+
+# Unit test / coverage reports
+htmlcov/
+.tox/
+.nox/
+.coverage
+.coverage.*
+.cache
+nosetests.xml
+coverage.xml
+*.cover
+*.py,cover
+.hypothesis/
+.pytest_cache/
+cover/
+
+# Translations
+*.mo
+*.pot
+
+# Django stuff:
+*.log
+local_settings.py
+db.sqlite3
+db.sqlite3-journal
+
+# Flask stuff:
+instance/
+.webassets-cache
+
+# Scrapy stuff:
+.scrapy
+
+# Sphinx documentation
+docs/_build/
+
+# PyBuilder
+.pybuilder/
+target/
+
+# Jupyter Notebook
+.ipynb_checkpoints
+
+# IPython
+profile_default/
+ipython_config.py
+
+# pyenv
+#   For a library or package, you might want to ignore these files since the code is
+#   intended to run in multiple environments; otherwise, check them in:
+# .python-version
+
+# pipenv
+#   According to pypa/pipenv#598, it is recommended to include Pipfile.lock in version control.
+#   However, in case of collaboration, if having platform-specific dependencies or dependencies
+#   having no cross-platform support, pipenv may install dependencies that don't work, or not
+#   install all needed dependencies.
+#Pipfile.lock
+
+# UV
+#   Similar to Pipfile.lock, it is generally recommended to include uv.lock in version control.
+#   This is especially recommended for binary packages to ensure reproducibility, and is more
+#   commonly ignored for libraries.
+#uv.lock
+
+# poetry
+#   Similar to Pipfile.lock, it is generally recommended to include poetry.lock in version control.
+#   This is especially recommended for binary packages to ensure reproducibility, and is more
+#   commonly ignored for libraries.
+#   https://python-poetry.org/docs/basic-usage/#commit-your-poetrylock-file-to-version-control
+#poetry.lock
+
+# pdm
+#   Similar to Pipfile.lock, it is generally recommended to include pdm.lock in version control.
+#pdm.lock
+#   pdm stores project-wide configurations in .pdm.toml, but it is recommended to not include it
+#   in version control.
+#   https://pdm.fming.dev/latest/usage/project/#working-with-version-control
+.pdm.toml
+.pdm-python
+.pdm-build/
+
+# PEP 582; used by e.g. github.com/David-OConnor/pyflow and github.com/pdm-project/pdm
+__pypackages__/
+
+# Celery stuff
+celerybeat-schedule
+celerybeat.pid
+
+# SageMath parsed files
+*.sage.py
+
+# Environments
+.env
+.venv
+env/
+venv/
+ENV/
+env.bak/
+venv.bak/
+
+# Spyder project settings
+.spyderproject
+.spyproject
+
+# Rope project settings
+.ropeproject
+
+# mkdocs documentation
+/site
+
+# mypy
+.mypy_cache/
+.dmypy.json
+dmypy.json
+
+# Pyre type checker
+.pyre/
+
+# pytype static type analyzer
+.pytype/
+
+# Cython debug symbols
+cython_debug/
+
+# PyCharm
+#  JetBrains specific template is maintained in a separate JetBrains.gitignore that can
+#  be found at https://github.com/github/gitignore/blob/main/Global/JetBrains.gitignore
+#  and can be added to the global gitignore or merged into this file.  For a more nuclear
+#  option (not recommended) you can uncomment the following to ignore the entire idea folder.
+#.idea/
+
+# PyPI configuration file
+.pypirc

LICENSE

@@ -1,27 +1,24 @@
-Copyright (c) 2022, Max Bain
-All rights reserved.
+BSD 2-Clause License
+
+Copyright (c) 2024, Max Bain
 
 Redistribution and use in source and binary forms, with or without
 modification, are permitted provided that the following conditions are met:
-1. Redistributions of source code must retain the above copyright
-   notice, this list of conditions and the following disclaimer.
-2. Redistributions in binary form must reproduce the above copyright
-   notice, this list of conditions and the following disclaimer in the
-   documentation and/or other materials provided with the distribution.
-3. All advertising materials mentioning features or use of this software
-   must display the following acknowledgement:
-   This product includes software developed by Max Bain.
-4. Neither the name of Max Bain nor the
-   names of its contributors may be used to endorse or promote products
-   derived from this software without specific prior written permission.
 
-THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDER ''AS IS'' AND ANY
-EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED
-WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE
+1. Redistributions of source code must retain the above copyright notice, this
+   list of conditions and the following disclaimer.
+
+2. Redistributions in binary form must reproduce the above copyright notice,
+   this list of conditions and the following disclaimer in the documentation
+   and/or other materials provided with the distribution.
+
+THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS"
+AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
+IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE
 DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT HOLDER OR CONTRIBUTORS BE LIABLE
-FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 
-DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR 
-SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER 
+FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
+DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR
+SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER
 CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
-OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE 
-USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.

MANIFEST.in

@@ -1,4 +1,3 @@
 include whisperx/assets/*
-include whisperx/assets/gpt2/*
-include whisperx/assets/multilingual/*
-include whisperx/normalizers/english.json
+include LICENSE
+include requirements.txt

README.md

@@ -13,85 +13,106 @@
         <img src="https://img.shields.io/github/license/m-bain/whisperX.svg"
              alt="GitHub license">
   </a>
+  <a href="https://arxiv.org/abs/2303.00747">
+        <img src="http://img.shields.io/badge/Arxiv-2303.00747-B31B1B.svg"
+             alt="ArXiv paper">
+  </a>
   <a href="https://twitter.com/intent/tweet?text=&url=https%3A%2F%2Fgithub.com%2Fm-bain%2FwhisperX">
   <img src="https://img.shields.io/twitter/url/https/github.com/m-bain/whisperX.svg?style=social" alt="Twitter">
   </a>      
 </p>
 
-<p align="center">
-  <a href="#what-is-it">What is it</a> •
-  <a href="#setup">Setup</a> •
-  <a href="#example">Usage</a> •
-  <a href="#other-languages">Multilingual</a> •
-  <a href="#contribute">Contribute</a> •
-  <a href="EXAMPLES.md">More examples</a>
-</p>
+<img width="1216" align="center" alt="whisperx-arch" src="https://raw.githubusercontent.com/m-bain/whisperX/refs/heads/main/figures/pipeline.png">
 
+<!-- <p align="left">Whisper-Based Automatic Speech Recognition (ASR) with improved timestamp accuracy + quality via forced phoneme alignment and voice-activity based batching for fast inference.</p> -->
 
-<img width="1216" align="center" alt="whisperx-arch" src="https://user-images.githubusercontent.com/36994049/211200186-8b779e26-0bfd-4127-aee2-5a9238b95e1f.png">
+<!-- <h2 align="left", id="what-is-it">What is it 🔎</h2> -->
 
+This repository provides fast automatic speech recognition (70x realtime with large-v2) with word-level timestamps and speaker diarization.
 
-<p align="left">Whisper-Based Automatic Speech Recognition (ASR) with improved timestamp accuracy using forced alignment.
+- ⚡️ Batched inference for 70x realtime transcription using whisper large-v2
+- 🪶 [faster-whisper](https://github.com/guillaumekln/faster-whisper) backend, requires <8GB gpu memory for large-v2 with beam_size=5
+- 🎯 Accurate word-level timestamps using wav2vec2 alignment
+- 👯‍♂️ Multispeaker ASR using speaker diarization from [pyannote-audio](https://github.com/pyannote/pyannote-audio) (speaker ID labels)
+- 🗣️ VAD preprocessing, reduces hallucination & batching with no WER degradation
 
-</p>
-
-
-<h2 align="left", id="what-is-it">What is it 🔎</h2>
-
-This repository refines the timestamps of openAI's Whisper model via forced aligment with phoneme-based ASR models (e.g. wav2vec2.0), multilingual use-case.
-
-
-**Whisper** is an ASR model [developed by OpenAI](https://github.com/openai/whisper), trained on a large dataset of diverse audio. Whilst it does produces highly accurate transcriptions, the corresponding timestamps are at the utterance-level, not per word, and can be inaccurate by several seconds.
+**Whisper** is an ASR model [developed by OpenAI](https://github.com/openai/whisper), trained on a large dataset of diverse audio. Whilst it does produces highly accurate transcriptions, the corresponding timestamps are at the utterance-level, not per word, and can be inaccurate by several seconds. OpenAI's whisper does not natively support batching.
 
 **Phoneme-Based ASR** A suite of models finetuned to recognise the smallest unit of speech distinguishing one word from another, e.g. the element p in "tap". A popular example model is [wav2vec2.0](https://huggingface.co/facebook/wav2vec2-large-960h-lv60-self).
 
 **Forced Alignment** refers to the process by which orthographic transcriptions are aligned to audio recordings to automatically generate phone level segmentation.
 
+**Voice Activity Detection (VAD)** is the detection of the presence or absence of human speech.
+
+**Speaker Diarization** is the process of partitioning an audio stream containing human speech into homogeneous segments according to the identity of each speaker.
+
 <h2 align="left", id="highlights">New🚨</h2>
 
-- Batch processing: Add `--vad_filter --parallel_bs [int]` for transcribing long audio file in batches (only supported with VAD filtering). Replace `[int]` with a batch size that fits your GPU memory, e.g. `--parallel_bs 16`.
-- VAD filtering: Voice Activity Detection (VAD) from [Pyannote.audio](https://huggingface.co/pyannote/voice-activity-detection) is used as a preprocessing step to remove reliance on whisper timestamps and only transcribe audio segments containing speech. add `--vad_filter` flag, increases timestamp accuracy and robustness (requires more GPU mem due to 30s inputs in wav2vec2)
-- Character level timestamps (see `*.char.ass` file output)
-- Diarization (still in beta, add `--diarize`)
-
+- 1st place at [Ego4d transcription challenge](https://eval.ai/web/challenges/challenge-page/1637/leaderboard/3931/WER) 🏆
+- _WhisperX_ accepted at INTERSPEECH 2023
+- v3 transcript segment-per-sentence: using nltk sent_tokenize for better subtitlting & better diarization
+- v3 released, 70x speed-up open-sourced. Using batched whisper with [faster-whisper](https://github.com/guillaumekln/faster-whisper) backend!
+- v2 released, code cleanup, imports whisper library VAD filtering is now turned on by default, as in the paper.
+- Paper drop🎓👨‍🏫! Please see our [ArxiV preprint](https://arxiv.org/abs/2303.00747) for benchmarking and details of WhisperX. We also introduce more efficient batch inference resulting in large-v2 with \*60-70x REAL TIME speed.
 
 <h2 align="left" id="setup">Setup ⚙️</h2>
-Install this package using
 
-`pip install git+https://github.com/m-bain/whisperx.git`
+### 1. Simple Installation (Recommended)
 
-If already installed, update package to most recent commit
+The easiest way to install WhisperX is through PyPi:
 
-`pip install git+https://github.com/m-bain/whisperx.git --upgrade`
-
-If wishing to modify this package, clone and install in editable mode:
+```bash
+pip install whisperx
 ```
-$ git clone https://github.com/m-bain/whisperX.git
-$ cd whisperX
-$ pip install -e .
+
+Or if using [uvx](https://docs.astral.sh/uv/guides/tools/#running-tools):
+
+```bash
+uvx whisperx
 ```
 
+### 2. Advanced Installation Options
+
+These installation methods are for developers or users with specific needs. If you're not sure, stick with the simple installation above.
+
+#### Option A: Install from GitHub
+
+To install directly from the GitHub repository:
+
+```bash
+uvx git+https://github.com/m-bain/whisperX.git
+```
+
+#### Option B: Developer Installation
+
+If you want to modify the code or contribute to the project:
+
+```bash
+git clone https://github.com/m-bain/whisperX.git
+cd whisperX
+uv sync --all-extras --dev
+```
+
+> **Note**: The development version may contain experimental features and bugs. Use the stable PyPI release for production environments.
+
 You may also need to install ffmpeg, rust etc. Follow openAI instructions here https://github.com/openai/whisper#setup.
 
+### Speaker Diarization
 
-### Voice Activity Detection Filtering & Diarization
-To **enable VAD filtering and Diarization**, include your Hugging Face access token that you can generate from [Here](https://huggingface.co/settings/tokens) after the `--hf_token` argument and accept the user agreement for the following models: [Segmentation](https://huggingface.co/pyannote/segmentation) , [Voice Activity Detection (VAD)](https://huggingface.co/pyannote/voice-activity-detection) , and [Speaker Diarization](https://huggingface.co/pyannote/speaker-diarization)
+To **enable Speaker Diarization**, include your Hugging Face access token (read) that you can generate from [Here](https://huggingface.co/settings/tokens) after the `--hf_token` argument and accept the user agreement for the following models: [Segmentation](https://huggingface.co/pyannote/segmentation-3.0) and [Speaker-Diarization-3.1](https://huggingface.co/pyannote/speaker-diarization-3.1) (if you choose to use Speaker-Diarization 2.x, follow requirements [here](https://huggingface.co/pyannote/speaker-diarization) instead.)
 
+> **Note**<br>
+> As of Oct 11, 2023, there is a known issue regarding slow performance with pyannote/Speaker-Diarization-3.0 in whisperX. It is due to dependency conflicts between faster-whisper and pyannote-audio 3.0.0. Please see [this issue](https://github.com/m-bain/whisperX/issues/499) for more details and potential workarounds.
 
 <h2 align="left" id="example">Usage 💬 (command line)</h2>
 
 ### English
 
-Run whisper on example segment (using default params)
+Run whisper on example segment (using default params, whisper small) add `--highlight_words True` to visualise word timings in the .srt file.
 
-    whisperx examples/sample01.wav
+    whisperx path/to/audio.wav
 
-
-For increased timestamp accuracy, at the cost of higher gpu mem, use bigger models and VAD filtering e.g.
-
-    whisperx examples/sample01.wav --model large-v2 --vad_filter --align_model WAV2VEC2_ASR_LARGE_LV60K_960H
-
-Result using *WhisperX* with forced alignment to wav2vec2.0 large:
+Result using _WhisperX_ with forced alignment to wav2vec2.0 large:
 
 https://user-images.githubusercontent.com/36994049/208253969-7e35fe2a-7541-434a-ae91-8e919540555d.mp4
 
@@ -99,140 +120,179 @@ Compare this to original whisper out the box, where many transcriptions are out
 
 https://user-images.githubusercontent.com/36994049/207743923-b4f0d537-29ae-4be2-b404-bb941db73652.mov
 
+For increased timestamp accuracy, at the cost of higher gpu mem, use bigger models (bigger alignment model not found to be that helpful, see paper) e.g.
+
+    whisperx path/to/audio.wav --model large-v2 --align_model WAV2VEC2_ASR_LARGE_LV60K_960H --batch_size 4
+
+To label the transcript with speaker ID's (set number of speakers if known e.g. `--min_speakers 2` `--max_speakers 2`):
+
+    whisperx path/to/audio.wav --model large-v2 --diarize --highlight_words True
+
+To run on CPU instead of GPU (and for running on Mac OS X):
+
+    whisperx path/to/audio.wav --compute_type int8
+
 ### Other languages
 
-The phoneme ASR alignment model is *language-specific*, for tested languages these models are [automatically picked from torchaudio pipelines or huggingface](https://github.com/m-bain/whisperX/blob/e909f2f766b23b2000f2d95df41f9b844ac53e49/whisperx/transcribe.py#L22).
+The phoneme ASR alignment model is _language-specific_, for tested languages these models are [automatically picked from torchaudio pipelines or huggingface](https://github.com/m-bain/whisperX/blob/f2da2f858e99e4211fe4f64b5f2938b007827e17/whisperx/alignment.py#L24-L58).
 Just pass in the `--language` code, and use the whisper `--model large`.
 
-Currently default models provided for `{en, fr, de, es, it, ja, zh, nl, uk, pt}`. If the detected language is not in this list, you need to find a phoneme-based ASR model from [huggingface model hub](https://huggingface.co/models) and test it on your data.
-
+Currently default models provided for `{en, fr, de, es, it}` via torchaudio pipelines and many other languages via Hugging Face. Please find the list of currently supported languages under `DEFAULT_ALIGN_MODELS_HF` on [alignment.py](https://github.com/m-bain/whisperX/blob/main/whisperx/alignment.py). If the detected language is not in this list, you need to find a phoneme-based ASR model from [huggingface model hub](https://huggingface.co/models) and test it on your data.
 
 #### E.g. German
-    whisperx --model large --language de examples/sample_de_01.wav
+
+    whisperx --model large-v2 --language de path/to/audio.wav
 
 https://user-images.githubusercontent.com/36994049/208298811-e36002ba-3698-4731-97d4-0aebd07e0eb3.mov
 
-
 See more examples in other languages [here](EXAMPLES.md).
 
-## Python usage  🐍
+## Python usage 🐍
 
 ```python
 import whisperx
+import gc
 
-device = "cuda" 
+device = "cuda"
 audio_file = "audio.mp3"
+batch_size = 16 # reduce if low on GPU mem
+compute_type = "float16" # change to "int8" if low on GPU mem (may reduce accuracy)
 
-# transcribe with original whisper
-model = whisperx.load_model("large", device)
-result = model.transcribe(audio_file)
+# 1. Transcribe with original whisper (batched)
+model = whisperx.load_model("large-v2", device, compute_type=compute_type)
 
+# save model to local path (optional)
+# model_dir = "/path/"
+# model = whisperx.load_model("large-v2", device, compute_type=compute_type, download_root=model_dir)
+
+audio = whisperx.load_audio(audio_file)
+result = model.transcribe(audio, batch_size=batch_size)
 print(result["segments"]) # before alignment
 
-# load alignment model and metadata
+# delete model if low on GPU resources
+# import gc; import torch; gc.collect(); torch.cuda.empty_cache(); del model
+
+# 2. Align whisper output
 model_a, metadata = whisperx.load_align_model(language_code=result["language"], device=device)
+result = whisperx.align(result["segments"], model_a, metadata, audio, device, return_char_alignments=False)
 
-# align whisper output
-result_aligned = whisperx.align(result["segments"], model_a, metadata, audio_file, device)
+print(result["segments"]) # after alignment
 
-print(result_aligned["segments"]) # after alignment
-print(result_aligned["word_segments"]) # after alignment
+# delete model if low on GPU resources
+# import gc; import torch; gc.collect(); torch.cuda.empty_cache(); del model_a
+
+# 3. Assign speaker labels
+diarize_model = whisperx.diarize.DiarizationPipeline(use_auth_token=YOUR_HF_TOKEN, device=device)
+
+# add min/max number of speakers if known
+diarize_segments = diarize_model(audio)
+# diarize_model(audio, min_speakers=min_speakers, max_speakers=max_speakers)
+
+result = whisperx.assign_word_speakers(diarize_segments, result)
+print(diarize_segments)
+print(result["segments"]) # segments are now assigned speaker IDs
 ```
 
+## Demos 🚀
 
-<h2 align="left" id="whisper-mod">Whisper Modifications</h2>
+[![Replicate (large-v3](https://img.shields.io/static/v1?label=Replicate+WhisperX+large-v3&message=Demo+%26+Cloud+API&color=blue)](https://replicate.com/victor-upmeet/whisperx)
+[![Replicate (large-v2](https://img.shields.io/static/v1?label=Replicate+WhisperX+large-v2&message=Demo+%26+Cloud+API&color=blue)](https://replicate.com/daanelson/whisperx)
+[![Replicate (medium)](https://img.shields.io/static/v1?label=Replicate+WhisperX+medium&message=Demo+%26+Cloud+API&color=blue)](https://replicate.com/carnifexer/whisperx)
 
-In addition to forced alignment, the following two modifications have been made to the whisper transcription method:
+If you don't have access to your own GPUs, use the links above to try out WhisperX.
 
-1. `--condition_on_prev_text` is set to `False` by default (reduces hallucination)
+<h2 align="left" id="whisper-mod">Technical Details 👷‍♂️</h2>
 
-2. Clamping segment `end_time` to be at least 0.02s (one time precision) later than `start_time` (prevents segments with negative duration)
+For specific details on the batching and alignment, the effect of VAD, as well as the chosen alignment model, see the preprint [paper](https://www.robots.ox.ac.uk/~vgg/publications/2023/Bain23/bain23.pdf).
 
+To reduce GPU memory requirements, try any of the following (2. & 3. can affect quality):
+
+1.  reduce batch size, e.g. `--batch_size 4`
+2.  use a smaller ASR model `--model base`
+3.  Use lighter compute type `--compute_type int8`
+
+Transcription differences from openai's whisper:
+
+1. Transcription without timestamps. To enable single pass batching, whisper inference is performed `--without_timestamps True`, this ensures 1 forward pass per sample in the batch. However, this can cause discrepancies the default whisper output.
+2. VAD-based segment transcription, unlike the buffered transcription of openai's. In the WhisperX paper we show this reduces WER, and enables accurate batched inference
+3. `--condition_on_prev_text` is set to `False` by default (reduces hallucination)
 
 <h2 align="left" id="limitations">Limitations ⚠️</h2>
 
-- Not thoroughly tested, especially for non-english, results may vary -- please post issue to let me know the results on your data
-- Whisper normalises spoken numbers e.g. "fifty seven" to arabic numerals "57". Need to perform this normalization after alignment, so the phonemes can be aligned. Currently just ignores numbers.
-- If not using VAD filter, whisperx assumes the initial whisper timestamps are accurate to some degree (within margin of 2 seconds, adjust if needed -- bigger margins more prone to alignment errors)
+- Transcript words which do not contain characters in the alignment models dictionary e.g. "2014." or "£13.60" cannot be aligned and therefore are not given a timing.
 - Overlapping speech is not handled particularly well by whisper nor whisperx
-- Diariazation is far from perfect.
-
+- Diarization is far from perfect
+- Language specific wav2vec2 model is needed
 
 <h2 align="left" id="contribute">Contribute 🧑‍🏫</h2>
 
-If you are multilingual, a major way you can contribute to this project is to find phoneme models on huggingface (or train your own) and test them on speech for the target language. If the results look good send a merge request and some examples showing its success.
+If you are multilingual, a major way you can contribute to this project is to find phoneme models on huggingface (or train your own) and test them on speech for the target language. If the results look good send a pull request and some examples showing its success.
 
-The next major upgrade we are working on is whisper with speaker diarization, so if you have any experience on this please share.
+Bug finding and pull requests are also highly appreciated to keep this project going, since it's already diverging from the original research scope.
 
-<h2 align="left" id="coming-soon">Coming Soon 🗓</h2>
+<h2 align="left" id="coming-soon">TODO 🗓</h2>
 
-* [x] Multilingual init
+- [x] Multilingual init
 
-* [x] Subtitle .ass output
+- [x] Automatic align model selection based on language detection
 
-* [x] Automatic align model selection based on language detection
+- [x] Python usage
 
-* [x] Python usage
+- [x] Incorporating speaker diarization
 
-* [x] Character level timestamps
+- [x] Model flush, for low gpu mem resources
 
-* [x] Incorporating  speaker diarization
+- [x] Faster-whisper backend
 
-* [x] Inference speedup with batch processing
+- [x] Add max-line etc. see (openai's whisper utils.py)
 
-* [ ] Improve diarization (word level). *Harder than first thought...*
+- [x] Sentence-level segments (nltk toolbox)
 
+- [x] Improve alignment logic
+
+- [ ] update examples with diarization and word highlighting
+
+- [ ] Subtitle .ass output <- bring this back (removed in v3)
+
+- [ ] Add benchmarking code (TEDLIUM for spd/WER & word segmentation)
+
+- [x] Allow silero-vad as alternative VAD option
+
+- [ ] Improve diarization (word level). _Harder than first thought..._
 
 <h2 align="left" id="contact">Contact/Support 📇</h2>
 
-Contact maxbain[at]robots[dot]ox[dot]ac[dot]uk for queries
+Contact maxhbain@gmail.com for queries.
 
 <a href="https://www.buymeacoffee.com/maxhbain" target="_blank"><img src="https://cdn.buymeacoffee.com/buttons/default-orange.png" alt="Buy Me A Coffee" height="41" width="174"></a>
 
-
 <h2 align="left" id="acks">Acknowledgements 🙏</h2>
 
-This work, and my PhD, is supported by the [VGG (Visual Geometry Group)](https://www.robots.ox.ac.uk/~vgg/) and University of Oxford.
-
-
+This work, and my PhD, is supported by the [VGG (Visual Geometry Group)](https://www.robots.ox.ac.uk/~vgg/) and the University of Oxford.
 
 Of course, this is builds on [openAI's whisper](https://github.com/openai/whisper).
-And borrows important alignment code from [PyTorch tutorial on forced alignment](https://pytorch.org/tutorials/intermediate/forced_alignment_with_torchaudio_tutorial.html)
+Borrows important alignment code from [PyTorch tutorial on forced alignment](https://pytorch.org/tutorials/intermediate/forced_alignment_with_torchaudio_tutorial.html)
+And uses the wonderful pyannote VAD / Diarization https://github.com/pyannote/pyannote-audio
 
+Valuable VAD & Diarization Models from:
+
+- [pyannote audio][https://github.com/pyannote/pyannote-audio]
+- [silero vad][https://github.com/snakers4/silero-vad]
+
+Great backend from [faster-whisper](https://github.com/guillaumekln/faster-whisper) and [CTranslate2](https://github.com/OpenNMT/CTranslate2)
+
+Those who have [supported this work financially](https://www.buymeacoffee.com/maxhbain) 🙏
+
+Finally, thanks to the OS [contributors](https://github.com/m-bain/whisperX/graphs/contributors) of this project, keeping it going and identifying bugs.
 
 <h2 align="left" id="cite">Citation</h2>
-If you use this in your research, for now just cite the repo,
+If you use this in your research, please cite the paper:
 
 ```bibtex
-@misc{bain2022whisperx,
-  author = {Bain, Max and Han, Tengda},
-  title = {WhisperX},
-  year = {2022},
-  publisher = {GitHub},
-  journal = {GitHub repository},
-  howpublished = {\url{https://github.com/m-bain/whisperX}},
-}
-```
-
-as well as the whisper paper,
-
-```bibtex
-@article{radford2022robust,
-  title={Robust speech recognition via large-scale weak supervision},
-  author={Radford, Alec and Kim, Jong Wook and Xu, Tao and Brockman, Greg and McLeavey, Christine and Sutskever, Ilya},
-  journal={arXiv preprint arXiv:2212.04356},
-  year={2022}
-}
-```
-and any alignment model used, e.g. wav2vec2.0.
-
-```bibtex
-@article{baevski2020wav2vec,
-  title={wav2vec 2.0: A framework for self-supervised learning of speech representations},
-  author={Baevski, Alexei and Zhou, Yuhao and Mohamed, Abdelrahman and Auli, Michael},
-  journal={Advances in Neural Information Processing Systems},
-  volume={33},
-  pages={12449--12460},
-  year={2020}
+@article{bain2022whisperx,
+  title={WhisperX: Time-Accurate Speech Transcription of Long-Form Audio},
+  author={Bain, Max and Huh, Jaesung and Han, Tengda and Zisserman, Andrew},
+  journal={INTERSPEECH 2023},
+  year={2023}
 }
 ```

examples/whisper/sample01.wav.srt

@@ -1,140 +0,0 @@
-1
-00:00:00,000 --> 00:00:03,000
-Bella, Gloria, love.
-
-2
-00:00:03,000 --> 00:00:04,000
-Oh.
-
-3
-00:00:04,000 --> 00:00:05,000
-How are you?
-
-4
-00:00:05,000 --> 00:00:07,000
-Oh, I'm OK.
-
-5
-00:00:07,000 --> 00:00:08,000
-I will be.
-
-6
-00:00:08,000 --> 00:00:09,000
-I said she could stay with us tomorrow
-
-7
-00:00:09,000 --> 00:00:10,000
-just until she feels better.
-
-8
-00:00:10,000 --> 00:00:11,000
-Yeah.
-
-9
-00:00:11,000 --> 00:00:12,000
-Of course she can.
-
-10
-00:00:12,000 --> 00:00:14,000
-No, things won't be for long.
-
-11
-00:00:14,000 --> 00:00:16,000
-Well, you can stay as long as you want, my love.
-
-12
-00:00:16,000 --> 00:00:18,000
-I've really missed you.
-
-13
-00:00:18,000 --> 00:00:19,000
-Pops.
-
-14
-00:00:19,000 --> 00:00:20,000
-Great to see you, love.
-
-15
-00:00:20,000 --> 00:00:22,000
-Oh.
-
-16
-00:00:22,000 --> 00:00:23,000
-All right, shall we get you off to bed then?
-
-17
-00:00:23,000 --> 00:00:25,000
-You should have given me some warm.
-
-18
-00:00:25,000 --> 00:00:26,000
-I know.
-
-19
-00:00:26,000 --> 00:00:27,000
-I'll have to put the electric blanket on.
-
-20
-00:00:27,000 --> 00:00:28,000
-I'm sorry.
-
-21
-00:00:28,000 --> 00:00:29,000
-All right, Bella.
-
-22
-00:00:29,000 --> 00:00:31,000
-Freezing up there.
-
-23
-00:00:31,000 --> 00:00:34,000
-In a bedroom, Peter unpacks her suitcase.
-
-24
-00:00:34,000 --> 00:00:38,000
-The middle-aged woman opens her green case.
-
-25
-00:00:38,000 --> 00:00:39,000
-Do you want your PJs?
-
-26
-00:00:39,000 --> 00:00:40,000
-Yeah.
-
-27
-00:00:40,000 --> 00:00:42,000
-Yeah.
-
-28
-00:00:42,000 --> 00:00:45,000
-Lifting a bundle of pajamas, Peter finds a sheet of paper
-
-29
-00:00:45,000 --> 00:00:50,000
-labeled Lancaster North Hospital discharge sheet.
-
-30
-00:00:50,000 --> 00:00:52,000
-He closes the suitcase and brings Gloria the pajamas.
-
-31
-00:00:52,000 --> 00:00:54,000
-There you go.
-
-32
-00:00:54,000 --> 00:00:55,000
-Thank you.
-
-33
-00:00:55,000 --> 00:00:57,000
-He picks up the locket.
-
-34
-00:00:57,000 --> 00:00:59,000
-He kept it.
-
-35
-00:00:59,000 --> 00:01:28,000
-Oh, cool.
-

examples/whisper/sample_fr_01.wav.srt

@@ -1,92 +0,0 @@
-1
-00:00:00,000 --> 00:00:01,240
-Lâchez, c'est bon.
-
-2
-00:00:01,240 --> 00:00:02,240
-Ça va?
-
-3
-00:00:02,240 --> 00:00:03,240
-Oui.
-
-4
-00:00:03,240 --> 00:00:04,240
-Merci beaucoup.
-
-5
-00:00:04,240 --> 00:00:05,240
-Chèque ou espèce?
-
-6
-00:00:05,240 --> 00:00:08,640
-J'ai un chèque sur la commode, il est signé.
-
-7
-00:00:08,640 --> 00:00:09,640
-Je vais le repirer.
-
-8
-00:00:09,640 --> 00:00:10,640
-Ok.
-
-9
-00:00:10,640 --> 00:00:11,640
-Ouh là!
-
-10
-00:00:11,640 --> 00:00:12,640
-Venez.
-
-11
-00:00:12,640 --> 00:00:13,640
-Merci.
-
-12
-00:00:13,640 --> 00:00:14,640
-Ah! C'est qui?
-
-13
-00:00:14,640 --> 00:00:21,640
-C'est pas vrai, qu'est-ce qu'il fout ici, ce con?
-
-14
-00:00:21,640 --> 00:00:26,640
-Excusez-moi, mais je crois que j'ai oublié mon sac chez vous.
-
-15
-00:00:26,640 --> 00:00:27,640
-Ça va?
-
-16
-00:00:27,640 --> 00:00:44,200
-Attendez, tout à l'heure là, c'était vous? Vous? Pas lui? Vous?
-
-17
-00:00:44,200 --> 00:00:48,360
-Vous avez tout à fait raison, M. Xanaquis, Malek est à l'interne brillant qui apprend
-
-18
-00:00:48,360 --> 00:00:49,360
-le métier avec moi.
-
-19
-00:00:49,360 --> 00:00:50,360
-Ah!
-
-20
-00:00:50,360 --> 00:00:51,360
-Bien.
-
-21
-00:00:51,360 --> 00:00:55,520
-Justement, il y a la famille Boboune qui m'attend pour une consultation.
-
-22
-00:00:55,520 --> 00:00:56,520
-Qui?
-
-23
-00:00:56,520 --> 00:00:57,760
-Faisons pas attendre les Boboune, allez.
-

examples/whisper/sample_fr_01.wav.txt

@@ -1,23 +0,0 @@
-Lâchez, c'est bon.
-Ça va?
-Oui.
-Merci beaucoup.
-Chèque ou espèce?
-J'ai un chèque sur la commode, il est signé.
-Je vais le repirer.
-Ok.
-Ouh là!
-Venez.
-Merci.
-Ah! C'est qui?
-C'est pas vrai, qu'est-ce qu'il fout ici, ce con?
-Excusez-moi, mais je crois que j'ai oublié mon sac chez vous.
-Ça va?
-Attendez, tout à l'heure là, c'était vous? Vous? Pas lui? Vous?
-Vous avez tout à fait raison, M. Xanaquis, Malek est à l'interne brillant qui apprend
-le métier avec moi.
-Ah!
-Bien.
-Justement, il y a la famille Boboune qui m'attend pour une consultation.
-Qui?
-Faisons pas attendre les Boboune, allez.

examples/whisper/sample_fr_01.wav.vtt

@@ -1,71 +0,0 @@
-WEBVTT
-
-00:00.000 --> 00:01.240
-Lâchez, c'est bon.
-
-00:01.240 --> 00:02.240
-Ça va?
-
-00:02.240 --> 00:03.240
-Oui.
-
-00:03.240 --> 00:04.240
-Merci beaucoup.
-
-00:04.240 --> 00:05.240
-Chèque ou espèce?
-
-00:05.240 --> 00:08.640
-J'ai un chèque sur la commode, il est signé.
-
-00:08.640 --> 00:09.640
-Je vais le repirer.
-
-00:09.640 --> 00:10.640
-Ok.
-
-00:10.640 --> 00:11.640
-Ouh là!
-
-00:11.640 --> 00:12.640
-Venez.
-
-00:12.640 --> 00:13.640
-Merci.
-
-00:13.640 --> 00:14.640
-Ah! C'est qui?
-
-00:14.640 --> 00:21.640
-C'est pas vrai, qu'est-ce qu'il fout ici, ce con?
-
-00:21.640 --> 00:26.640
-Excusez-moi, mais je crois que j'ai oublié mon sac chez vous.
-
-00:26.640 --> 00:27.640
-Ça va?
-
-00:27.640 --> 00:44.200
-Attendez, tout à l'heure là, c'était vous? Vous? Pas lui? Vous?
-
-00:44.200 --> 00:48.360
-Vous avez tout à fait raison, M. Xanaquis, Malek est à l'interne brillant qui apprend
-
-00:48.360 --> 00:49.360
-le métier avec moi.
-
-00:49.360 --> 00:50.360
-Ah!
-
-00:50.360 --> 00:51.360
-Bien.
-
-00:51.360 --> 00:55.520
-Justement, il y a la famille Boboune qui m'attend pour une consultation.
-
-00:55.520 --> 00:56.520
-Qui?
-
-00:56.520 --> 00:57.760
-Faisons pas attendre les Boboune, allez.
-

examples/whisperx/sample01.wav.ass

@@ -1,290 +0,0 @@
-[Script Info]
-ScriptType: v4.00+
-PlayResX: 384
-PlayResY: 288
-ScaledBorderAndShadow: yes
-
-[V4+ Styles]
-Format: Name, Fontname, Fontsize, PrimaryColour, SecondaryColour, OutlineColour, BackColour, Bold, Italic, Underline, StrikeOut, ScaleX, ScaleY, Spacing, Angle, BorderStyle, Outline, Shadow, Alignment, MarginL, MarginR, MarginV, Encoding
-Style: Default,Arial,24,&Hffffff,&Hffffff,&H0,&H0,0,0,0,0,100,100,0,0,1,1,0,2,10,10,10,0
-
-[Events]
-Format: Layer, Start, End, Style, Name, MarginL, MarginR, MarginV, Effect, Text
-
-Dialogue: 0,0:00:1.18,0:00:1.67,Default,,0,0,0,,{\1c&HFF00&\u1}Bella,{\r} Gloria, love.
-Dialogue: 0,0:00:1.67,0:00:2.65,Default,,0,0,0,,Bella, Gloria, love.
-Dialogue: 0,0:00:2.65,0:00:3.05,Default,,0,0,0,,Bella, {\1c&HFF00&\u1}Gloria,{\r} love.
-Dialogue: 0,0:00:3.05,0:00:3.07,Default,,0,0,0,,Bella, Gloria, love.
-Dialogue: 0,0:00:3.07,0:00:3.27,Default,,0,0,0,,Bella, Gloria, {\1c&HFF00&\u1}love.{\r}
-Dialogue: 0,0:00:3.75,0:00:3.85,Default,,0,0,0,,{\1c&HFF00&\u1}Oh.{\r}
-Dialogue: 0,0:00:4.50,0:00:4.72,Default,,0,0,0,,{\1c&HFF00&\u1}How{\r} are you?
-Dialogue: 0,0:00:4.72,0:00:5.78,Default,,0,0,0,,How are you?
-Dialogue: 0,0:00:5.78,0:00:5.90,Default,,0,0,0,,How {\1c&HFF00&\u1}are{\r} you?
-Dialogue: 0,0:00:5.90,0:00:5.94,Default,,0,0,0,,How are you?
-Dialogue: 0,0:00:5.94,0:00:6.22,Default,,0,0,0,,How are {\1c&HFF00&\u1}you?{\r}
-Dialogue: 0,0:00:6.72,0:00:6.80,Default,,0,0,0,,{\1c&HFF00&\u1}Oh,{\r} I'm OK.
-Dialogue: 0,0:00:6.80,0:00:6.88,Default,,0,0,0,,Oh, I'm OK.
-Dialogue: 0,0:00:6.88,0:00:7.04,Default,,0,0,0,,Oh, {\1c&HFF00&\u1}I'm{\r} OK.
-Dialogue: 0,0:00:7.04,0:00:7.09,Default,,0,0,0,,Oh, I'm OK.
-Dialogue: 0,0:00:7.09,0:00:7.13,Default,,0,0,0,,Oh, I'm {\1c&HFF00&\u1}OK.{\r}
-Dialogue: 0,0:00:8.41,0:00:8.45,Default,,0,0,0,,{\1c&HFF00&\u1}I{\r} will be.
-Dialogue: 0,0:00:8.45,0:00:8.49,Default,,0,0,0,,I will be.
-Dialogue: 0,0:00:8.49,0:00:8.73,Default,,0,0,0,,I {\1c&HFF00&\u1}will{\r} be.
-Dialogue: 0,0:00:8.73,0:00:8.77,Default,,0,0,0,,I will be.
-Dialogue: 0,0:00:8.77,0:00:8.91,Default,,0,0,0,,I will {\1c&HFF00&\u1}be.{\r}
-Dialogue: 0,0:00:9.22,0:00:9.30,Default,,0,0,0,,{\1c&HFF00&\u1}I{\r} said she could stay with us tomorrow
-Dialogue: 0,0:00:9.30,0:00:9.34,Default,,0,0,0,,I said she could stay with us tomorrow
-Dialogue: 0,0:00:9.34,0:00:9.48,Default,,0,0,0,,I {\1c&HFF00&\u1}said{\r} she could stay with us tomorrow
-Dialogue: 0,0:00:9.48,0:00:9.52,Default,,0,0,0,,I said she could stay with us tomorrow
-Dialogue: 0,0:00:9.52,0:00:9.60,Default,,0,0,0,,I said {\1c&HFF00&\u1}she{\r} could stay with us tomorrow
-Dialogue: 0,0:00:9.60,0:00:9.62,Default,,0,0,0,,I said she could stay with us tomorrow
-Dialogue: 0,0:00:9.62,0:00:9.76,Default,,0,0,0,,I said she {\1c&HFF00&\u1}could{\r} stay with us tomorrow
-Dialogue: 0,0:00:9.76,0:00:9.78,Default,,0,0,0,,I said she could stay with us tomorrow
-Dialogue: 0,0:00:9.78,0:00:9.94,Default,,0,0,0,,I said she could {\1c&HFF00&\u1}stay{\r} with us tomorrow
-Dialogue: 0,0:00:9.94,0:00:9.96,Default,,0,0,0,,I said she could stay with us tomorrow
-Dialogue: 0,0:00:9.96,0:00:10.08,Default,,0,0,0,,I said she could stay {\1c&HFF00&\u1}with{\r} us tomorrow
-Dialogue: 0,0:00:10.08,0:00:10.10,Default,,0,0,0,,I said she could stay with us tomorrow
-Dialogue: 0,0:00:10.10,0:00:10.14,Default,,0,0,0,,I said she could stay with {\1c&HFF00&\u1}us{\r} tomorrow
-Dialogue: 0,0:00:10.14,0:00:10.16,Default,,0,0,0,,I said she could stay with us tomorrow
-Dialogue: 0,0:00:10.16,0:00:10.44,Default,,0,0,0,,I said she could stay with us {\1c&HFF00&\u1}tomorrow{\r}
-Dialogue: 0,0:00:10.46,0:00:10.54,Default,,0,0,0,,{\1c&HFF00&\u1}just{\r} until she feels better.
-Dialogue: 0,0:00:10.54,0:00:10.56,Default,,0,0,0,,just until she feels better.
-Dialogue: 0,0:00:10.56,0:00:10.68,Default,,0,0,0,,just {\1c&HFF00&\u1}until{\r} she feels better.
-Dialogue: 0,0:00:10.68,0:00:10.70,Default,,0,0,0,,just until she feels better.
-Dialogue: 0,0:00:10.70,0:00:10.80,Default,,0,0,0,,just until {\1c&HFF00&\u1}she{\r} feels better.
-Dialogue: 0,0:00:10.80,0:00:10.82,Default,,0,0,0,,just until she feels better.
-Dialogue: 0,0:00:10.82,0:00:11.05,Default,,0,0,0,,just until she {\1c&HFF00&\u1}feels{\r} better.
-Dialogue: 0,0:00:11.05,0:00:11.09,Default,,0,0,0,,just until she feels better.
-Dialogue: 0,0:00:11.09,0:00:11.35,Default,,0,0,0,,just until she feels {\1c&HFF00&\u1}better.{\r}
-Dialogue: 0,0:00:11.73,0:00:11.95,Default,,0,0,0,,{\1c&HFF00&\u1}Yeah.{\r}
-Dialogue: 0,0:00:12.09,0:00:12.17,Default,,0,0,0,,{\1c&HFF00&\u1}Of{\r} course she can.
-Dialogue: 0,0:00:12.17,0:00:12.20,Default,,0,0,0,,Of course she can.
-Dialogue: 0,0:00:12.20,0:00:12.32,Default,,0,0,0,,Of {\1c&HFF00&\u1}course{\r} she can.
-Dialogue: 0,0:00:12.32,0:00:12.38,Default,,0,0,0,,Of course she can.
-Dialogue: 0,0:00:12.38,0:00:12.64,Default,,0,0,0,,Of course {\1c&HFF00&\u1}she{\r} can.
-Dialogue: 0,0:00:12.64,0:00:12.72,Default,,0,0,0,,Of course she can.
-Dialogue: 0,0:00:12.72,0:00:13.24,Default,,0,0,0,,Of course she {\1c&HFF00&\u1}can.{\r}
-Dialogue: 0,0:00:13.36,0:00:13.70,Default,,0,0,0,,{\1c&HFF00&\u1}No,{\r} things won't be for long.
-Dialogue: 0,0:00:13.70,0:00:13.82,Default,,0,0,0,,No, things won't be for long.
-Dialogue: 0,0:00:13.82,0:00:14.12,Default,,0,0,0,,No, {\1c&HFF00&\u1}things{\r} won't be for long.
-Dialogue: 0,0:00:14.12,0:00:14.19,Default,,0,0,0,,No, things won't be for long.
-Dialogue: 0,0:00:14.19,0:00:14.39,Default,,0,0,0,,No, things {\1c&HFF00&\u1}won't{\r} be for long.
-Dialogue: 0,0:00:14.39,0:00:14.43,Default,,0,0,0,,No, things won't be for long.
-Dialogue: 0,0:00:14.43,0:00:14.53,Default,,0,0,0,,No, things won't {\1c&HFF00&\u1}be{\r} for long.
-Dialogue: 0,0:00:14.53,0:00:14.59,Default,,0,0,0,,No, things won't be for long.
-Dialogue: 0,0:00:14.59,0:00:14.73,Default,,0,0,0,,No, things won't be {\1c&HFF00&\u1}for{\r} long.
-Dialogue: 0,0:00:14.73,0:00:14.81,Default,,0,0,0,,No, things won't be for long.
-Dialogue: 0,0:00:14.81,0:00:15.01,Default,,0,0,0,,No, things won't be for {\1c&HFF00&\u1}long.{\r}
-Dialogue: 0,0:00:15.17,0:00:15.41,Default,,0,0,0,,{\1c&HFF00&\u1}Well,{\r} you can stay as long as you want, my love.
-Dialogue: 0,0:00:15.41,0:00:15.43,Default,,0,0,0,,Well, you can stay as long as you want, my love.
-Dialogue: 0,0:00:15.43,0:00:15.51,Default,,0,0,0,,Well, {\1c&HFF00&\u1}you{\r} can stay as long as you want, my love.
-Dialogue: 0,0:00:15.51,0:00:15.55,Default,,0,0,0,,Well, you can stay as long as you want, my love.
-Dialogue: 0,0:00:15.55,0:00:15.69,Default,,0,0,0,,Well, you {\1c&HFF00&\u1}can{\r} stay as long as you want, my love.
-Dialogue: 0,0:00:15.69,0:00:15.75,Default,,0,0,0,,Well, you can stay as long as you want, my love.
-Dialogue: 0,0:00:15.75,0:00:15.89,Default,,0,0,0,,Well, you can {\1c&HFF00&\u1}stay{\r} as long as you want, my love.
-Dialogue: 0,0:00:15.89,0:00:15.95,Default,,0,0,0,,Well, you can stay as long as you want, my love.
-Dialogue: 0,0:00:15.95,0:00:16.01,Default,,0,0,0,,Well, you can stay {\1c&HFF00&\u1}as{\r} long as you want, my love.
-Dialogue: 0,0:00:16.01,0:00:16.05,Default,,0,0,0,,Well, you can stay as long as you want, my love.
-Dialogue: 0,0:00:16.05,0:00:16.20,Default,,0,0,0,,Well, you can stay as {\1c&HFF00&\u1}long{\r} as you want, my love.
-Dialogue: 0,0:00:16.20,0:00:16.24,Default,,0,0,0,,Well, you can stay as long as you want, my love.
-Dialogue: 0,0:00:16.24,0:00:16.28,Default,,0,0,0,,Well, you can stay as long {\1c&HFF00&\u1}as{\r} you want, my love.
-Dialogue: 0,0:00:16.28,0:00:16.30,Default,,0,0,0,,Well, you can stay as long as you want, my love.
-Dialogue: 0,0:00:16.30,0:00:16.42,Default,,0,0,0,,Well, you can stay as long as {\1c&HFF00&\u1}you{\r} want, my love.
-Dialogue: 0,0:00:16.42,0:00:16.46,Default,,0,0,0,,Well, you can stay as long as you want, my love.
-Dialogue: 0,0:00:16.46,0:00:16.66,Default,,0,0,0,,Well, you can stay as long as you {\1c&HFF00&\u1}want,{\r} my love.
-Dialogue: 0,0:00:16.66,0:00:16.72,Default,,0,0,0,,Well, you can stay as long as you want, my love.
-Dialogue: 0,0:00:16.72,0:00:16.92,Default,,0,0,0,,Well, you can stay as long as you want, {\1c&HFF00&\u1}my{\r} love.
-Dialogue: 0,0:00:16.92,0:00:16.96,Default,,0,0,0,,Well, you can stay as long as you want, my love.
-Dialogue: 0,0:00:16.96,0:00:17.34,Default,,0,0,0,,Well, you can stay as long as you want, my {\1c&HFF00&\u1}love.{\r}
-Dialogue: 0,0:00:17.62,0:00:17.86,Default,,0,0,0,,{\1c&HFF00&\u1}I've{\r} really missed you.
-Dialogue: 0,0:00:17.86,0:00:17.88,Default,,0,0,0,,I've really missed you.
-Dialogue: 0,0:00:17.88,0:00:18.14,Default,,0,0,0,,I've {\1c&HFF00&\u1}really{\r} missed you.
-Dialogue: 0,0:00:18.14,0:00:18.19,Default,,0,0,0,,I've really missed you.
-Dialogue: 0,0:00:18.19,0:00:18.59,Default,,0,0,0,,I've really {\1c&HFF00&\u1}missed{\r} you.
-Dialogue: 0,0:00:18.59,0:00:18.63,Default,,0,0,0,,I've really missed you.
-Dialogue: 0,0:00:18.63,0:00:18.81,Default,,0,0,0,,I've really missed {\1c&HFF00&\u1}you.{\r}
-Dialogue: 0,0:00:19.49,0:00:19.79,Default,,0,0,0,,{\1c&HFF00&\u1}Pops.{\r}
-Dialogue: 0,0:00:20.40,0:00:20.64,Default,,0,0,0,,{\1c&HFF00&\u1}Great{\r} to see you, love.
-Dialogue: 0,0:00:20.64,0:00:20.66,Default,,0,0,0,,Great to see you, love.
-Dialogue: 0,0:00:20.66,0:00:20.78,Default,,0,0,0,,Great {\1c&HFF00&\u1}to{\r} see you, love.
-Dialogue: 0,0:00:20.78,0:00:20.82,Default,,0,0,0,,Great to see you, love.
-Dialogue: 0,0:00:20.82,0:00:21.14,Default,,0,0,0,,Great to {\1c&HFF00&\u1}see{\r} you, love.
-Dialogue: 0,0:00:21.14,0:00:21.16,Default,,0,0,0,,Great to see you, love.
-Dialogue: 0,0:00:21.16,0:00:21.28,Default,,0,0,0,,Great to see {\1c&HFF00&\u1}you,{\r} love.
-Dialogue: 0,0:00:21.28,0:00:21.32,Default,,0,0,0,,Great to see you, love.
-Dialogue: 0,0:00:21.32,0:00:21.68,Default,,0,0,0,,Great to see you, {\1c&HFF00&\u1}love.{\r}
-Dialogue: 0,0:00:21.90,0:00:23.21,Default,,0,0,0,,{\1c&HFF00&\u1}Oh.{\r}
-Dialogue: 0,0:00:23.23,0:00:23.29,Default,,0,0,0,,{\1c&HFF00&\u1}All{\r} right, shall we get you off to bed then?
-Dialogue: 0,0:00:23.29,0:00:23.31,Default,,0,0,0,,All right, shall we get you off to bed then?
-Dialogue: 0,0:00:23.31,0:00:23.41,Default,,0,0,0,,All {\1c&HFF00&\u1}right,{\r} shall we get you off to bed then?
-Dialogue: 0,0:00:23.41,0:00:23.43,Default,,0,0,0,,All right, shall we get you off to bed then?
-Dialogue: 0,0:00:23.43,0:00:23.55,Default,,0,0,0,,All right, {\1c&HFF00&\u1}shall{\r} we get you off to bed then?
-Dialogue: 0,0:00:23.55,0:00:23.57,Default,,0,0,0,,All right, shall we get you off to bed then?
-Dialogue: 0,0:00:23.57,0:00:23.65,Default,,0,0,0,,All right, shall {\1c&HFF00&\u1}we{\r} get you off to bed then?
-Dialogue: 0,0:00:23.65,0:00:23.67,Default,,0,0,0,,All right, shall we get you off to bed then?
-Dialogue: 0,0:00:23.67,0:00:23.74,Default,,0,0,0,,All right, shall we {\1c&HFF00&\u1}get{\r} you off to bed then?
-Dialogue: 0,0:00:23.74,0:00:23.76,Default,,0,0,0,,All right, shall we get you off to bed then?
-Dialogue: 0,0:00:23.76,0:00:23.82,Default,,0,0,0,,All right, shall we get {\1c&HFF00&\u1}you{\r} off to bed then?
-Dialogue: 0,0:00:23.82,0:00:23.84,Default,,0,0,0,,All right, shall we get you off to bed then?
-Dialogue: 0,0:00:23.84,0:00:23.94,Default,,0,0,0,,All right, shall we get you {\1c&HFF00&\u1}off{\r} to bed then?
-Dialogue: 0,0:00:23.94,0:00:23.96,Default,,0,0,0,,All right, shall we get you off to bed then?
-Dialogue: 0,0:00:23.96,0:00:24.04,Default,,0,0,0,,All right, shall we get you off {\1c&HFF00&\u1}to{\r} bed then?
-Dialogue: 0,0:00:24.04,0:00:24.06,Default,,0,0,0,,All right, shall we get you off to bed then?
-Dialogue: 0,0:00:24.06,0:00:24.22,Default,,0,0,0,,All right, shall we get you off to {\1c&HFF00&\u1}bed{\r} then?
-Dialogue: 0,0:00:24.22,0:00:24.24,Default,,0,0,0,,All right, shall we get you off to bed then?
-Dialogue: 0,0:00:24.24,0:00:24.38,Default,,0,0,0,,All right, shall we get you off to bed {\1c&HFF00&\u1}then?{\r}
-Dialogue: 0,0:00:24.58,0:00:24.72,Default,,0,0,0,,{\1c&HFF00&\u1}You{\r} should have given me some warm.
-Dialogue: 0,0:00:24.72,0:00:24.78,Default,,0,0,0,,You should have given me some warm.
-Dialogue: 0,0:00:24.78,0:00:24.98,Default,,0,0,0,,You {\1c&HFF00&\u1}should{\r} have given me some warm.
-Dialogue: 0,0:00:24.98,0:00:25.02,Default,,0,0,0,,You should have given me some warm.
-Dialogue: 0,0:00:25.02,0:00:25.12,Default,,0,0,0,,You should {\1c&HFF00&\u1}have{\r} given me some warm.
-Dialogue: 0,0:00:25.12,0:00:25.16,Default,,0,0,0,,You should have given me some warm.
-Dialogue: 0,0:00:25.16,0:00:25.29,Default,,0,0,0,,You should have {\1c&HFF00&\u1}given{\r} me some warm.
-Dialogue: 0,0:00:25.29,0:00:25.35,Default,,0,0,0,,You should have given me some warm.
-Dialogue: 0,0:00:25.35,0:00:25.45,Default,,0,0,0,,You should have given {\1c&HFF00&\u1}me{\r} some warm.
-Dialogue: 0,0:00:25.45,0:00:25.49,Default,,0,0,0,,You should have given me some warm.
-Dialogue: 0,0:00:25.49,0:00:25.67,Default,,0,0,0,,You should have given me {\1c&HFF00&\u1}some{\r} warm.
-Dialogue: 0,0:00:25.67,0:00:25.81,Default,,0,0,0,,You should have given me some warm.
-Dialogue: 0,0:00:25.81,0:00:26.05,Default,,0,0,0,,You should have given me some {\1c&HFF00&\u1}warm.{\r}
-Dialogue: 0,0:00:26.31,0:00:26.37,Default,,0,0,0,,{\1c&HFF00&\u1}I{\r} know.
-Dialogue: 0,0:00:26.37,0:00:26.39,Default,,0,0,0,,I know.
-Dialogue: 0,0:00:26.39,0:00:26.49,Default,,0,0,0,,I {\1c&HFF00&\u1}know.{\r}
-Dialogue: 0,0:00:26.61,0:00:26.69,Default,,0,0,0,,{\1c&HFF00&\u1}I'll{\r} have to put the electric blanket on.
-Dialogue: 0,0:00:26.69,0:00:26.71,Default,,0,0,0,,I'll have to put the electric blanket on.
-Dialogue: 0,0:00:26.71,0:00:26.81,Default,,0,0,0,,I'll {\1c&HFF00&\u1}have{\r} to put the electric blanket on.
-Dialogue: 0,0:00:26.81,0:00:26.83,Default,,0,0,0,,I'll have to put the electric blanket on.
-Dialogue: 0,0:00:26.83,0:00:27.02,Default,,0,0,0,,I'll have {\1c&HFF00&\u1}to{\r} put the electric blanket on.
-Dialogue: 0,0:00:27.02,0:00:27.06,Default,,0,0,0,,I'll have to put the electric blanket on.
-Dialogue: 0,0:00:27.06,0:00:27.42,Default,,0,0,0,,I'll have to {\1c&HFF00&\u1}put{\r} the electric blanket on.
-Dialogue: 0,0:00:27.42,0:00:27.48,Default,,0,0,0,,I'll have to put the electric blanket on.
-Dialogue: 0,0:00:27.48,0:00:27.56,Default,,0,0,0,,I'll have to put {\1c&HFF00&\u1}the{\r} electric blanket on.
-Dialogue: 0,0:00:27.56,0:00:27.70,Default,,0,0,0,,I'll have to put the electric blanket on.
-Dialogue: 0,0:00:27.70,0:00:28.08,Default,,0,0,0,,I'll have to put the {\1c&HFF00&\u1}electric{\r} blanket on.
-Dialogue: 0,0:00:28.08,0:00:28.62,Default,,0,0,0,,I'll have to put the electric blanket on.
-Dialogue: 0,0:00:28.62,0:00:28.84,Default,,0,0,0,,I'll have to put the electric {\1c&HFF00&\u1}blanket{\r} on.
-Dialogue: 0,0:00:28.84,0:00:28.90,Default,,0,0,0,,I'll have to put the electric blanket on.
-Dialogue: 0,0:00:28.90,0:00:28.94,Default,,0,0,0,,I'll have to put the electric blanket {\1c&HFF00&\u1}on.{\r}
-Dialogue: 0,0:00:29.49,0:00:29.55,Default,,0,0,0,,{\1c&HFF00&\u1}I'm{\r} sorry.
-Dialogue: 0,0:00:29.55,0:00:29.57,Default,,0,0,0,,I'm sorry.
-Dialogue: 0,0:00:29.57,0:00:29.82,Default,,0,0,0,,I'm {\1c&HFF00&\u1}sorry.{\r}
-Dialogue: 0,0:00:29.98,0:00:30.08,Default,,0,0,0,,{\1c&HFF00&\u1}All{\r} right, Bella.
-Dialogue: 0,0:00:30.08,0:00:30.10,Default,,0,0,0,,All right, Bella.
-Dialogue: 0,0:00:30.10,0:00:30.29,Default,,0,0,0,,All {\1c&HFF00&\u1}right,{\r} Bella.
-Dialogue: 0,0:00:30.29,0:00:30.43,Default,,0,0,0,,All right, Bella.
-Dialogue: 0,0:00:30.43,0:00:30.63,Default,,0,0,0,,All right, {\1c&HFF00&\u1}Bella.{\r}
-Dialogue: 0,0:00:31.37,0:00:31.58,Default,,0,0,0,,{\1c&HFF00&\u1}Freezing{\r} up there.
-Dialogue: 0,0:00:31.58,0:00:31.62,Default,,0,0,0,,Freezing up there.
-Dialogue: 0,0:00:31.62,0:00:31.66,Default,,0,0,0,,Freezing {\1c&HFF00&\u1}up{\r} there.
-Dialogue: 0,0:00:31.66,0:00:31.68,Default,,0,0,0,,Freezing up there.
-Dialogue: 0,0:00:31.68,0:00:31.90,Default,,0,0,0,,Freezing up {\1c&HFF00&\u1}there.{\r}
-Dialogue: 0,0:00:31.90,0:00:31.94,Default,,0,0,0,,{\1c&HFF00&\u1}In{\r} a bedroom, Peter unpacks her suitcase.
-Dialogue: 0,0:00:31.94,0:00:31.96,Default,,0,0,0,,In a bedroom, Peter unpacks her suitcase.
-Dialogue: 0,0:00:31.96,0:00:31.98,Default,,0,0,0,,In {\1c&HFF00&\u1}a{\r} bedroom, Peter unpacks her suitcase.
-Dialogue: 0,0:00:31.98,0:00:32.00,Default,,0,0,0,,In a bedroom, Peter unpacks her suitcase.
-Dialogue: 0,0:00:32.00,0:00:32.14,Default,,0,0,0,,In a {\1c&HFF00&\u1}bedroom,{\r} Peter unpacks her suitcase.
-Dialogue: 0,0:00:32.14,0:00:32.20,Default,,0,0,0,,In a bedroom, Peter unpacks her suitcase.
-Dialogue: 0,0:00:32.20,0:00:32.50,Default,,0,0,0,,In a bedroom, {\1c&HFF00&\u1}Peter{\r} unpacks her suitcase.
-Dialogue: 0,0:00:32.50,0:00:32.58,Default,,0,0,0,,In a bedroom, Peter unpacks her suitcase.
-Dialogue: 0,0:00:32.58,0:00:32.98,Default,,0,0,0,,In a bedroom, Peter {\1c&HFF00&\u1}unpacks{\r} her suitcase.
-Dialogue: 0,0:00:32.98,0:00:33.00,Default,,0,0,0,,In a bedroom, Peter unpacks her suitcase.
-Dialogue: 0,0:00:33.00,0:00:33.10,Default,,0,0,0,,In a bedroom, Peter unpacks {\1c&HFF00&\u1}her{\r} suitcase.
-Dialogue: 0,0:00:33.10,0:00:33.16,Default,,0,0,0,,In a bedroom, Peter unpacks her suitcase.
-Dialogue: 0,0:00:33.16,0:00:33.65,Default,,0,0,0,,In a bedroom, Peter unpacks her {\1c&HFF00&\u1}suitcase.{\r}
-Dialogue: 0,0:00:34.27,0:00:34.35,Default,,0,0,0,,{\1c&HFF00&\u1}The{\r} middle-aged woman opens her green case.
-Dialogue: 0,0:00:34.35,0:00:34.39,Default,,0,0,0,,The middle-aged woman opens her green case.
-Dialogue: 0,0:00:34.39,0:00:34.91,Default,,0,0,0,,The {\1c&HFF00&\u1}middle-aged{\r} woman opens her green case.
-Dialogue: 0,0:00:34.91,0:00:34.99,Default,,0,0,0,,The middle-aged woman opens her green case.
-Dialogue: 0,0:00:34.99,0:00:35.27,Default,,0,0,0,,The middle-aged {\1c&HFF00&\u1}woman{\r} opens her green case.
-Dialogue: 0,0:00:35.27,0:00:35.39,Default,,0,0,0,,The middle-aged woman opens her green case.
-Dialogue: 0,0:00:35.39,0:00:35.67,Default,,0,0,0,,The middle-aged woman {\1c&HFF00&\u1}opens{\r} her green case.
-Dialogue: 0,0:00:35.67,0:00:35.71,Default,,0,0,0,,The middle-aged woman opens her green case.
-Dialogue: 0,0:00:35.71,0:00:35.81,Default,,0,0,0,,The middle-aged woman opens {\1c&HFF00&\u1}her{\r} green case.
-Dialogue: 0,0:00:35.81,0:00:35.85,Default,,0,0,0,,The middle-aged woman opens her green case.
-Dialogue: 0,0:00:35.85,0:00:36.19,Default,,0,0,0,,The middle-aged woman opens her {\1c&HFF00&\u1}green{\r} case.
-Dialogue: 0,0:00:36.19,0:00:36.23,Default,,0,0,0,,The middle-aged woman opens her green case.
-Dialogue: 0,0:00:36.23,0:00:36.53,Default,,0,0,0,,The middle-aged woman opens her green {\1c&HFF00&\u1}case.{\r}
-Dialogue: 0,0:00:38.13,0:00:38.25,Default,,0,0,0,,{\1c&HFF00&\u1}Do{\r} you want your PJs?
-Dialogue: 0,0:00:38.25,0:00:38.28,Default,,0,0,0,,Do you want your PJs?
-Dialogue: 0,0:00:38.28,0:00:38.36,Default,,0,0,0,,Do {\1c&HFF00&\u1}you{\r} want your PJs?
-Dialogue: 0,0:00:38.36,0:00:38.38,Default,,0,0,0,,Do you want your PJs?
-Dialogue: 0,0:00:38.38,0:00:38.54,Default,,0,0,0,,Do you {\1c&HFF00&\u1}want{\r} your PJs?
-Dialogue: 0,0:00:38.54,0:00:38.56,Default,,0,0,0,,Do you want your PJs?
-Dialogue: 0,0:00:38.56,0:00:38.74,Default,,0,0,0,,Do you want {\1c&HFF00&\u1}your{\r} PJs?
-Dialogue: 0,0:00:38.74,0:00:38.88,Default,,0,0,0,,Do you want your PJs?
-Dialogue: 0,0:00:38.88,0:00:39.30,Default,,0,0,0,,Do you want your {\1c&HFF00&\u1}PJs?{\r}
-Dialogue: 0,0:00:39.88,0:00:40.18,Default,,0,0,0,,{\1c&HFF00&\u1}Yeah.{\r}
-Dialogue: 0,0:00:42.39,0:00:42.69,Default,,0,0,0,,{\1c&HFF00&\u1}Lifting{\r} a bundle of pajamas,
-Dialogue: 0,0:00:42.69,0:00:42.73,Default,,0,0,0,,Lifting a bundle of pajamas,
-Dialogue: 0,0:00:42.73,0:00:42.75,Default,,0,0,0,,Lifting {\1c&HFF00&\u1}a{\r} bundle of pajamas,
-Dialogue: 0,0:00:42.75,0:00:42.81,Default,,0,0,0,,Lifting a bundle of pajamas,
-Dialogue: 0,0:00:42.81,0:00:43.11,Default,,0,0,0,,Lifting a {\1c&HFF00&\u1}bundle{\r} of pajamas,
-Dialogue: 0,0:00:43.11,0:00:43.13,Default,,0,0,0,,Lifting a bundle of pajamas,
-Dialogue: 0,0:00:43.13,0:00:43.19,Default,,0,0,0,,Lifting a bundle {\1c&HFF00&\u1}of{\r} pajamas,
-Dialogue: 0,0:00:43.19,0:00:43.25,Default,,0,0,0,,Lifting a bundle of pajamas,
-Dialogue: 0,0:00:43.25,0:00:43.77,Default,,0,0,0,,Lifting a bundle of {\1c&HFF00&\u1}pajamas,{\r}
-Dialogue: 0,0:00:44.07,0:00:44.31,Default,,0,0,0,,{\1c&HFF00&\u1}Peter{\r} finds a sheet of paper labeled
-Dialogue: 0,0:00:44.31,0:00:44.37,Default,,0,0,0,,Peter finds a sheet of paper labeled
-Dialogue: 0,0:00:44.37,0:00:44.63,Default,,0,0,0,,Peter {\1c&HFF00&\u1}finds{\r} a sheet of paper labeled
-Dialogue: 0,0:00:44.63,0:00:44.67,Default,,0,0,0,,Peter finds a sheet of paper labeled
-Dialogue: 0,0:00:44.67,0:00:44.69,Default,,0,0,0,,Peter finds {\1c&HFF00&\u1}a{\r} sheet of paper labeled
-Dialogue: 0,0:00:44.69,0:00:44.75,Default,,0,0,0,,Peter finds a sheet of paper labeled
-Dialogue: 0,0:00:44.75,0:00:44.95,Default,,0,0,0,,Peter finds a {\1c&HFF00&\u1}sheet{\r} of paper labeled
-Dialogue: 0,0:00:44.95,0:00:44.99,Default,,0,0,0,,Peter finds a sheet of paper labeled
-Dialogue: 0,0:00:44.99,0:00:45.05,Default,,0,0,0,,Peter finds a sheet {\1c&HFF00&\u1}of{\r} paper labeled
-Dialogue: 0,0:00:45.05,0:00:45.11,Default,,0,0,0,,Peter finds a sheet of paper labeled
-Dialogue: 0,0:00:45.11,0:00:45.46,Default,,0,0,0,,Peter finds a sheet of {\1c&HFF00&\u1}paper{\r} labeled
-Dialogue: 0,0:00:45.46,0:00:45.54,Default,,0,0,0,,Peter finds a sheet of paper labeled
-Dialogue: 0,0:00:45.54,0:00:45.88,Default,,0,0,0,,Peter finds a sheet of paper {\1c&HFF00&\u1}labeled{\r}
-Dialogue: 0,0:00:46.34,0:00:47.04,Default,,0,0,0,,{\1c&HFF00&\u1}Lancaster{\r} North Hospital discharge sheet.
-Dialogue: 0,0:00:47.04,0:00:47.12,Default,,0,0,0,,Lancaster North Hospital discharge sheet.
-Dialogue: 0,0:00:47.12,0:00:47.38,Default,,0,0,0,,Lancaster {\1c&HFF00&\u1}North{\r} Hospital discharge sheet.
-Dialogue: 0,0:00:47.38,0:00:47.44,Default,,0,0,0,,Lancaster North Hospital discharge sheet.
-Dialogue: 0,0:00:47.44,0:00:47.94,Default,,0,0,0,,Lancaster North {\1c&HFF00&\u1}Hospital{\r} discharge sheet.
-Dialogue: 0,0:00:47.94,0:00:48.27,Default,,0,0,0,,Lancaster North Hospital discharge sheet.
-Dialogue: 0,0:00:48.27,0:00:48.93,Default,,0,0,0,,Lancaster North Hospital {\1c&HFF00&\u1}discharge{\r} sheet.
-Dialogue: 0,0:00:48.93,0:00:49.03,Default,,0,0,0,,Lancaster North Hospital discharge sheet.
-Dialogue: 0,0:00:49.03,0:00:49.25,Default,,0,0,0,,Lancaster North Hospital discharge {\1c&HFF00&\u1}sheet.{\r}
-Dialogue: 0,0:00:50.29,0:00:50.37,Default,,0,0,0,,{\1c&HFF00&\u1}He{\r} closes the suitcase and brings Gloria the pajamas.
-Dialogue: 0,0:00:50.37,0:00:50.41,Default,,0,0,0,,He closes the suitcase and brings Gloria the pajamas.
-Dialogue: 0,0:00:50.41,0:00:50.77,Default,,0,0,0,,He {\1c&HFF00&\u1}closes{\r} the suitcase and brings Gloria the pajamas.
-Dialogue: 0,0:00:50.77,0:00:50.81,Default,,0,0,0,,He closes the suitcase and brings Gloria the pajamas.
-Dialogue: 0,0:00:50.81,0:00:50.91,Default,,0,0,0,,He closes {\1c&HFF00&\u1}the{\r} suitcase and brings Gloria the pajamas.
-Dialogue: 0,0:00:50.91,0:00:50.95,Default,,0,0,0,,He closes the suitcase and brings Gloria the pajamas.
-Dialogue: 0,0:00:50.95,0:00:51.39,Default,,0,0,0,,He closes the {\1c&HFF00&\u1}suitcase{\r} and brings Gloria the pajamas.
-Dialogue: 0,0:00:51.39,0:00:51.43,Default,,0,0,0,,He closes the suitcase and brings Gloria the pajamas.
-Dialogue: 0,0:00:51.43,0:00:51.51,Default,,0,0,0,,He closes the suitcase {\1c&HFF00&\u1}and{\r} brings Gloria the pajamas.
-Dialogue: 0,0:00:51.51,0:00:51.53,Default,,0,0,0,,He closes the suitcase and brings Gloria the pajamas.
-Dialogue: 0,0:00:51.53,0:00:51.79,Default,,0,0,0,,He closes the suitcase and {\1c&HFF00&\u1}brings{\r} Gloria the pajamas.
-Dialogue: 0,0:00:51.79,0:00:51.83,Default,,0,0,0,,He closes the suitcase and brings Gloria the pajamas.
-Dialogue: 0,0:00:51.83,0:00:52.23,Default,,0,0,0,,He closes the suitcase and brings {\1c&HFF00&\u1}Gloria{\r} the pajamas.
-Dialogue: 0,0:00:52.23,0:00:52.25,Default,,0,0,0,,He closes the suitcase and brings Gloria the pajamas.
-Dialogue: 0,0:00:52.25,0:00:52.32,Default,,0,0,0,,He closes the suitcase and brings Gloria {\1c&HFF00&\u1}the{\r} pajamas.
-Dialogue: 0,0:00:52.32,0:00:52.36,Default,,0,0,0,,He closes the suitcase and brings Gloria the pajamas.
-Dialogue: 0,0:00:52.36,0:00:52.86,Default,,0,0,0,,He closes the suitcase and brings Gloria the {\1c&HFF00&\u1}pajamas.{\r}
-Dialogue: 0,0:00:54.19,0:00:54.49,Default,,0,0,0,,{\1c&HFF00&\u1}There{\r} you go.
-Dialogue: 0,0:00:54.49,0:00:54.55,Default,,0,0,0,,There you go.
-Dialogue: 0,0:00:54.55,0:00:54.77,Default,,0,0,0,,There {\1c&HFF00&\u1}you{\r} go.
-Dialogue: 0,0:00:54.77,0:00:54.79,Default,,0,0,0,,There you go.
-Dialogue: 0,0:00:54.79,0:00:54.83,Default,,0,0,0,,There you {\1c&HFF00&\u1}go.{\r}
-Dialogue: 0,0:00:55.65,0:00:55.77,Default,,0,0,0,,{\1c&HFF00&\u1}Thank{\r} you.
-Dialogue: 0,0:00:55.77,0:00:55.80,Default,,0,0,0,,Thank you.
-Dialogue: 0,0:00:55.80,0:00:55.90,Default,,0,0,0,,Thank {\1c&HFF00&\u1}you.{\r}
-Dialogue: 0,0:00:55.90,0:00:55.94,Default,,0,0,0,,{\1c&HFF00&\u1}He{\r} picks up the locket.
-Dialogue: 0,0:00:55.94,0:00:55.96,Default,,0,0,0,,He picks up the locket.
-Dialogue: 0,0:00:55.96,0:00:56.10,Default,,0,0,0,,He {\1c&HFF00&\u1}picks{\r} up the locket.
-Dialogue: 0,0:00:56.10,0:00:56.12,Default,,0,0,0,,He picks up the locket.
-Dialogue: 0,0:00:56.12,0:00:56.20,Default,,0,0,0,,He picks {\1c&HFF00&\u1}up{\r} the locket.
-Dialogue: 0,0:00:56.20,0:00:56.22,Default,,0,0,0,,He picks up the locket.
-Dialogue: 0,0:00:56.22,0:00:56.32,Default,,0,0,0,,He picks up {\1c&HFF00&\u1}the{\r} locket.
-Dialogue: 0,0:00:56.32,0:00:56.36,Default,,0,0,0,,He picks up the locket.
-Dialogue: 0,0:00:56.36,0:00:56.74,Default,,0,0,0,,He picks up the {\1c&HFF00&\u1}locket.{\r}
-Dialogue: 0,0:00:57.12,0:00:57.22,Default,,0,0,0,,{\1c&HFF00&\u1}You{\r} kept it.
-Dialogue: 0,0:00:57.22,0:00:57.27,Default,,0,0,0,,You kept it.
-Dialogue: 0,0:00:57.27,0:00:57.47,Default,,0,0,0,,You {\1c&HFF00&\u1}kept{\r} it.
-Dialogue: 0,0:00:57.47,0:00:57.55,Default,,0,0,0,,You kept it.
-Dialogue: 0,0:00:57.55,0:00:57.63,Default,,0,0,0,,You kept {\1c&HFF00&\u1}it.{\r}
-Dialogue: 0,0:00:58.87,0:00:58.99,Default,,0,0,0,,{\1c&HFF00&\u1}Oh,{\r} of course.
-Dialogue: 0,0:00:58.99,0:00:59.28,Default,,0,0,0,,Oh, of course.
-Dialogue: 0,0:00:59.28,0:00:59.58,Default,,0,0,0,,Oh, {\1c&HFF00&\u1}of{\r} course.
-Dialogue: 0,0:00:59.58,0:00:59.68,Default,,0,0,0,,Oh, of course.
-Dialogue: 0,0:00:59.68,0:00:59.96,Default,,0,0,0,,Oh, of {\1c&HFF00&\u1}course.{\r}

examples/whisperx/sample01.wav.srt

@@ -1,140 +0,0 @@
-1
-00:00:01,185 --> 00:00:03,273
-Bella, Gloria, love.
-
-2
-00:00:03,754 --> 00:00:03,855
-Oh.
-
-3
-00:00:04,496 --> 00:00:06,219
-How are you?
-
-4
-00:00:06,723 --> 00:00:07,126
-Oh, I'm OK.
-
-5
-00:00:08,412 --> 00:00:08,915
-I will be.
-
-6
-00:00:09,215 --> 00:00:10,439
-I said she could stay with us tomorrow
-
-7
-00:00:10,459 --> 00:00:11,351
-just until she feels better.
-
-8
-00:00:11,733 --> 00:00:11,954
-Yeah.
-
-9
-00:00:12,095 --> 00:00:13,238
-Of course she can.
-
-10
-00:00:13,359 --> 00:00:15,012
-No, things won't be for long.
-
-11
-00:00:15,173 --> 00:00:17,338
-Well, you can stay as long as you want, my love.
-
-12
-00:00:17,621 --> 00:00:18,810
-I've really missed you.
-
-13
-00:00:19,493 --> 00:00:19,795
-Pops.
-
-14
-00:00:20,396 --> 00:00:21,679
-Great to see you, love.
-
-15
-00:00:21,901 --> 00:00:23,213
-Oh.
-
-16
-00:00:23,233 --> 00:00:24,378
-All right, shall we get you off to bed then?
-
-17
-00:00:24,579 --> 00:00:26,052
-You should have given me some warm.
-
-18
-00:00:26,313 --> 00:00:26,494
-I know.
-
-19
-00:00:26,614 --> 00:00:28,940
-I'll have to put the electric blanket on.
-
-20
-00:00:29,490 --> 00:00:29,817
-I'm sorry.
-
-21
-00:00:29,980 --> 00:00:30,633
-All right, Bella.
-
-22
-00:00:31,375 --> 00:00:31,897
-Freezing up there.
-
-23
-00:00:31,897 --> 00:00:33,647
-In a bedroom, Peter unpacks her suitcase.
-
-24
-00:00:34,268 --> 00:00:36,533
-The middle-aged woman opens her green case.
-
-25
-00:00:38,135 --> 00:00:39,296
-Do you want your PJs?
-
-26
-00:00:39,879 --> 00:00:40,181
-Yeah.
-
-27
-00:00:42,388 --> 00:00:43,773
-Lifting a bundle of pajamas,
-
-28
-00:00:44,073 --> 00:00:45,876
-Peter finds a sheet of paper labeled
-
-29
-00:00:46,338 --> 00:00:49,249
-Lancaster North Hospital discharge sheet.
-
-30
-00:00:50,291 --> 00:00:52,856
-He closes the suitcase and brings Gloria the pajamas.
-
-31
-00:00:54,186 --> 00:00:54,831
-There you go.
-
-32
-00:00:55,654 --> 00:00:55,895
-Thank you.
-
-33
-00:00:55,895 --> 00:00:56,742
-He picks up the locket.
-
-34
-00:00:57,124 --> 00:00:57,627
-You kept it.
-
-35
-00:00:58,874 --> 00:00:59,960
-Oh, of course.
-

examples/whisperx/sample01.wav.word.srt

@@ -1,624 +0,0 @@
-1
-00:00:01,185 --> 00:00:01,667
-Bella,
-
-2
-00:00:02,651 --> 00:00:03,052
-Gloria,
-
-3
-00:00:03,072 --> 00:00:03,273
-love.
-
-4
-00:00:03,754 --> 00:00:03,855
-Oh.
-
-5
-00:00:04,496 --> 00:00:04,716
-How
-
-6
-00:00:05,778 --> 00:00:05,898
-are
-
-7
-00:00:05,938 --> 00:00:06,219
-you?
-
-8
-00:00:06,723 --> 00:00:06,803
-Oh,
-
-9
-00:00:06,884 --> 00:00:07,045
-I'm
-
-10
-00:00:07,085 --> 00:00:07,126
-OK.
-
-11
-00:00:08,412 --> 00:00:08,452
-I
-
-12
-00:00:08,492 --> 00:00:08,734
-will
-
-13
-00:00:08,774 --> 00:00:08,915
-be.
-
-14
-00:00:09,215 --> 00:00:09,296
-I
-
-15
-00:00:09,336 --> 00:00:09,476
-said
-
-16
-00:00:09,516 --> 00:00:09,596
-she
-
-17
-00:00:09,616 --> 00:00:09,757
-could
-
-18
-00:00:09,777 --> 00:00:09,937
-stay
-
-19
-00:00:09,957 --> 00:00:10,078
-with
-
-20
-00:00:10,098 --> 00:00:10,138
-us
-
-21
-00:00:10,158 --> 00:00:10,439
-tomorrow
-
-22
-00:00:10,459 --> 00:00:10,540
-just
-
-23
-00:00:10,560 --> 00:00:10,682
-until
-
-24
-00:00:10,702 --> 00:00:10,804
-she
-
-25
-00:00:10,824 --> 00:00:11,047
-feels
-
-26
-00:00:11,087 --> 00:00:11,351
-better.
-
-27
-00:00:11,733 --> 00:00:11,954
-Yeah.
-
-28
-00:00:12,095 --> 00:00:12,175
-Of
-
-29
-00:00:12,195 --> 00:00:12,315
-course
-
-30
-00:00:12,376 --> 00:00:12,636
-she
-
-31
-00:00:12,716 --> 00:00:13,238
-can.
-
-32
-00:00:13,359 --> 00:00:13,702
-No,
-
-33
-00:00:13,823 --> 00:00:14,125
-things
-
-34
-00:00:14,185 --> 00:00:14,387
-won't
-
-35
-00:00:14,427 --> 00:00:14,528
-be
-
-36
-00:00:14,589 --> 00:00:14,730
-for
-
-37
-00:00:14,810 --> 00:00:15,012
-long.
-
-38
-00:00:15,173 --> 00:00:15,413
-Well,
-
-39
-00:00:15,433 --> 00:00:15,513
-you
-
-40
-00:00:15,554 --> 00:00:15,694
-can
-
-41
-00:00:15,754 --> 00:00:15,894
-stay
-
-42
-00:00:15,955 --> 00:00:16,015
-as
-
-43
-00:00:16,055 --> 00:00:16,195
-long
-
-44
-00:00:16,235 --> 00:00:16,275
-as
-
-45
-00:00:16,295 --> 00:00:16,416
-you
-
-46
-00:00:16,456 --> 00:00:16,656
-want,
-
-47
-00:00:16,717 --> 00:00:16,917
-my
-
-48
-00:00:16,957 --> 00:00:17,338
-love.
-
-49
-00:00:17,621 --> 00:00:17,863
-I've
-
-50
-00:00:17,883 --> 00:00:18,145
-really
-
-51
-00:00:18,185 --> 00:00:18,588
-missed
-
-52
-00:00:18,629 --> 00:00:18,810
-you.
-
-53
-00:00:19,493 --> 00:00:19,795
-Pops.
-
-54
-00:00:20,396 --> 00:00:20,637
-Great
-
-55
-00:00:20,657 --> 00:00:20,777
-to
-
-56
-00:00:20,817 --> 00:00:21,138
-see
-
-57
-00:00:21,158 --> 00:00:21,278
-you,
-
-58
-00:00:21,318 --> 00:00:21,679
-love.
-
-59
-00:00:21,901 --> 00:00:23,213
-Oh.
-
-60
-00:00:23,233 --> 00:00:23,293
-All
-
-61
-00:00:23,313 --> 00:00:23,414
-right,
-
-62
-00:00:23,434 --> 00:00:23,554
-shall
-
-63
-00:00:23,574 --> 00:00:23,655
-we
-
-64
-00:00:23,675 --> 00:00:23,735
-get
-
-65
-00:00:23,755 --> 00:00:23,815
-you
-
-66
-00:00:23,835 --> 00:00:23,936
-off
-
-67
-00:00:23,956 --> 00:00:24,036
-to
-
-68
-00:00:24,056 --> 00:00:24,217
-bed
-
-69
-00:00:24,237 --> 00:00:24,378
-then?
-
-70
-00:00:24,579 --> 00:00:24,720
-You
-
-71
-00:00:24,781 --> 00:00:24,983
-should
-
-72
-00:00:25,023 --> 00:00:25,124
-have
-
-73
-00:00:25,164 --> 00:00:25,285
-given
-
-74
-00:00:25,346 --> 00:00:25,447
-me
-
-75
-00:00:25,487 --> 00:00:25,669
-some
-
-76
-00:00:25,810 --> 00:00:26,052
-warm.
-
-77
-00:00:26,313 --> 00:00:26,373
-I
-
-78
-00:00:26,393 --> 00:00:26,494
-know.
-
-79
-00:00:26,614 --> 00:00:26,694
-I'll
-
-80
-00:00:26,714 --> 00:00:26,815
-have
-
-81
-00:00:26,835 --> 00:00:27,015
-to
-
-82
-00:00:27,055 --> 00:00:27,416
-put
-
-83
-00:00:27,476 --> 00:00:27,556
-the
-
-84
-00:00:27,697 --> 00:00:28,078
-electric
-
-85
-00:00:28,619 --> 00:00:28,840
-blanket
-
-86
-00:00:28,900 --> 00:00:28,940
-on.
-
-87
-00:00:29,490 --> 00:00:29,551
-I'm
-
-88
-00:00:29,572 --> 00:00:29,817
-sorry.
-
-89
-00:00:29,980 --> 00:00:30,082
-All
-
-90
-00:00:30,102 --> 00:00:30,286
-right,
-
-91
-00:00:30,429 --> 00:00:30,633
-Bella.
-
-92
-00:00:31,375 --> 00:00:31,576
-Freezing
-
-93
-00:00:31,616 --> 00:00:31,656
-up
-
-94
-00:00:31,676 --> 00:00:31,897
-there.
-
-95
-00:00:31,897 --> 00:00:31,937
-In
-
-96
-00:00:31,957 --> 00:00:31,977
-a
-
-97
-00:00:31,997 --> 00:00:32,138
-bedroom,
-
-98
-00:00:32,198 --> 00:00:32,500
-Peter
-
-99
-00:00:32,581 --> 00:00:32,983
-unpacks
-
-100
-00:00:33,003 --> 00:00:33,103
-her
-
-101
-00:00:33,164 --> 00:00:33,647
-suitcase.
-
-102
-00:00:34,268 --> 00:00:34,348
-The
-
-103
-00:00:34,388 --> 00:00:34,909
-middle-aged
-
-104
-00:00:34,989 --> 00:00:35,270
-woman
-
-105
-00:00:35,390 --> 00:00:35,671
-opens
-
-106
-00:00:35,711 --> 00:00:35,811
-her
-
-107
-00:00:35,851 --> 00:00:36,192
-green
-
-108
-00:00:36,232 --> 00:00:36,533
-case.
-
-109
-00:00:38,135 --> 00:00:38,255
-Do
-
-110
-00:00:38,275 --> 00:00:38,355
-you
-
-111
-00:00:38,375 --> 00:00:38,535
-want
-
-112
-00:00:38,555 --> 00:00:38,736
-your
-
-113
-00:00:38,876 --> 00:00:39,296
-PJs?
-
-114
-00:00:39,879 --> 00:00:40,181
-Yeah.
-
-115
-00:00:42,388 --> 00:00:42,689
-Lifting
-
-116
-00:00:42,729 --> 00:00:42,749
-a
-
-117
-00:00:42,809 --> 00:00:43,110
-bundle
-
-118
-00:00:43,131 --> 00:00:43,191
-of
-
-119
-00:00:43,251 --> 00:00:43,773
-pajamas,
-
-120
-00:00:44,073 --> 00:00:44,314
-Peter
-
-121
-00:00:44,374 --> 00:00:44,634
-finds
-
-122
-00:00:44,674 --> 00:00:44,694
-a
-
-123
-00:00:44,754 --> 00:00:44,955
-sheet
-
-124
-00:00:44,995 --> 00:00:45,055
-of
-
-125
-00:00:45,115 --> 00:00:45,456
-paper
-
-126
-00:00:45,536 --> 00:00:45,876
-labeled
-
-127
-00:00:46,338 --> 00:00:47,041
-Lancaster
-
-128
-00:00:47,121 --> 00:00:47,382
-North
-
-129
-00:00:47,442 --> 00:00:47,944
-Hospital
-
-130
-00:00:48,266 --> 00:00:48,928
-discharge
-
-131
-00:00:49,029 --> 00:00:49,249
-sheet.
-
-132
-00:00:50,291 --> 00:00:50,371
-He
-
-133
-00:00:50,412 --> 00:00:50,772
-closes
-
-134
-00:00:50,812 --> 00:00:50,912
-the
-
-135
-00:00:50,953 --> 00:00:51,393
-suitcase
-
-136
-00:00:51,433 --> 00:00:51,514
-and
-
-137
-00:00:51,534 --> 00:00:51,794
-brings
-
-138
-00:00:51,834 --> 00:00:52,235
-Gloria
-
-139
-00:00:52,255 --> 00:00:52,315
-the
-
-140
-00:00:52,355 --> 00:00:52,856
-pajamas.
-
-141
-00:00:54,186 --> 00:00:54,488
-There
-
-142
-00:00:54,549 --> 00:00:54,771
-you
-
-143
-00:00:54,791 --> 00:00:54,831
-go.
-
-144
-00:00:55,654 --> 00:00:55,775
-Thank
-
-145
-00:00:55,795 --> 00:00:55,895
-you.
-
-146
-00:00:55,895 --> 00:00:55,936
-He
-
-147
-00:00:55,956 --> 00:00:56,097
-picks
-
-148
-00:00:56,117 --> 00:00:56,198
-up
-
-149
-00:00:56,218 --> 00:00:56,319
-the
-
-150
-00:00:56,359 --> 00:00:56,742
-locket.
-
-151
-00:00:57,124 --> 00:00:57,225
-You
-
-152
-00:00:57,265 --> 00:00:57,466
-kept
-
-153
-00:00:57,547 --> 00:00:57,627
-it.
-
-154
-00:00:58,874 --> 00:00:58,994
-Oh,
-
-155
-00:00:59,276 --> 00:00:59,578
-of
-
-156
-00:00:59,678 --> 00:00:59,960
-course.
-

examples/whisperx/sample_de_01.wav.ass

@@ -1,184 +0,0 @@
-[Script Info]
-ScriptType: v4.00+
-PlayResX: 384
-PlayResY: 288
-ScaledBorderAndShadow: yes
-
-[V4+ Styles]
-Format: Name, Fontname, Fontsize, PrimaryColour, SecondaryColour, OutlineColour, BackColour, Bold, Italic, Underline, StrikeOut, ScaleX, ScaleY, Spacing, Angle, BorderStyle, Outline, Shadow, Alignment, MarginL, MarginR, MarginV, Encoding
-Style: Default,Arial,24,&Hffffff,&Hffffff,&H0,&H0,0,0,0,0,100,100,0,0,1,1,0,2,10,10,10,0
-
-[Events]
-Format: Layer, Start, End, Style, Name, MarginL, MarginR, MarginV, Effect, Text
-
-Dialogue: 0,0:00:0.56,0:00:0.96,Default,,0,0,0,,{\1c&HFF00&\u1}Weinlein{\r} von Hammersmann
-Dialogue: 0,0:00:0.96,0:00:0.98,Default,,0,0,0,,Weinlein von Hammersmann
-Dialogue: 0,0:00:0.98,0:00:1.21,Default,,0,0,0,,Weinlein {\1c&HFF00&\u1}von{\r} Hammersmann
-Dialogue: 0,0:00:1.21,0:00:1.23,Default,,0,0,0,,Weinlein von Hammersmann
-Dialogue: 0,0:00:1.23,0:00:1.87,Default,,0,0,0,,Weinlein von {\1c&HFF00&\u1}Hammersmann{\r}
-Dialogue: 0,0:00:6.19,0:00:6.61,Default,,0,0,0,,{\1c&HFF00&\u1}Oberst{\r} Lande, es ist lange her
-Dialogue: 0,0:00:6.61,0:00:6.63,Default,,0,0,0,,Oberst Lande, es ist lange her
-Dialogue: 0,0:00:6.63,0:00:6.97,Default,,0,0,0,,Oberst {\1c&HFF00&\u1}Lande,{\r} es ist lange her
-Dialogue: 0,0:00:6.97,0:00:7.01,Default,,0,0,0,,Oberst Lande, es ist lange her
-Dialogue: 0,0:00:7.01,0:00:7.17,Default,,0,0,0,,Oberst Lande, {\1c&HFF00&\u1}es{\r} ist lange her
-Dialogue: 0,0:00:7.17,0:00:7.21,Default,,0,0,0,,Oberst Lande, es ist lange her
-Dialogue: 0,0:00:7.21,0:00:7.47,Default,,0,0,0,,Oberst Lande, es {\1c&HFF00&\u1}ist{\r} lange her
-Dialogue: 0,0:00:7.47,0:00:7.49,Default,,0,0,0,,Oberst Lande, es ist lange her
-Dialogue: 0,0:00:7.49,0:00:7.77,Default,,0,0,0,,Oberst Lande, es ist {\1c&HFF00&\u1}lange{\r} her
-Dialogue: 0,0:00:7.77,0:00:7.83,Default,,0,0,0,,Oberst Lande, es ist lange her
-Dialogue: 0,0:00:7.83,0:00:8.01,Default,,0,0,0,,Oberst Lande, es ist lange {\1c&HFF00&\u1}her{\r}
-Dialogue: 0,0:00:8.01,0:00:8.17,Default,,0,0,0,,{\1c&HFF00&\u1}Schneide{\r} ich wie eh und je
-Dialogue: 0,0:00:8.17,0:00:8.19,Default,,0,0,0,,Schneide ich wie eh und je
-Dialogue: 0,0:00:8.19,0:00:8.25,Default,,0,0,0,,Schneide {\1c&HFF00&\u1}ich{\r} wie eh und je
-Dialogue: 0,0:00:8.25,0:00:8.27,Default,,0,0,0,,Schneide ich wie eh und je
-Dialogue: 0,0:00:8.27,0:00:8.33,Default,,0,0,0,,Schneide ich {\1c&HFF00&\u1}wie{\r} eh und je
-Dialogue: 0,0:00:8.33,0:00:8.35,Default,,0,0,0,,Schneide ich wie eh und je
-Dialogue: 0,0:00:8.35,0:00:8.39,Default,,0,0,0,,Schneide ich wie {\1c&HFF00&\u1}eh{\r} und je
-Dialogue: 0,0:00:8.39,0:00:8.41,Default,,0,0,0,,Schneide ich wie eh und je
-Dialogue: 0,0:00:8.41,0:00:8.47,Default,,0,0,0,,Schneide ich wie eh {\1c&HFF00&\u1}und{\r} je
-Dialogue: 0,0:00:8.47,0:00:8.49,Default,,0,0,0,,Schneide ich wie eh und je
-Dialogue: 0,0:00:8.49,0:00:8.53,Default,,0,0,0,,Schneide ich wie eh und {\1c&HFF00&\u1}je{\r}
-Dialogue: 0,0:00:13.99,0:00:14.17,Default,,0,0,0,,{\1c&HFF00&\u1}Also{\r} was ist mit Ihrem wunderschönen Bein geschehen?
-Dialogue: 0,0:00:14.17,0:00:14.21,Default,,0,0,0,,Also was ist mit Ihrem wunderschönen Bein geschehen?
-Dialogue: 0,0:00:14.21,0:00:14.45,Default,,0,0,0,,Also {\1c&HFF00&\u1}was{\r} ist mit Ihrem wunderschönen Bein geschehen?
-Dialogue: 0,0:00:14.45,0:00:14.47,Default,,0,0,0,,Also was ist mit Ihrem wunderschönen Bein geschehen?
-Dialogue: 0,0:00:14.47,0:00:14.59,Default,,0,0,0,,Also was {\1c&HFF00&\u1}ist{\r} mit Ihrem wunderschönen Bein geschehen?
-Dialogue: 0,0:00:14.59,0:00:14.61,Default,,0,0,0,,Also was ist mit Ihrem wunderschönen Bein geschehen?
-Dialogue: 0,0:00:14.61,0:00:14.71,Default,,0,0,0,,Also was ist {\1c&HFF00&\u1}mit{\r} Ihrem wunderschönen Bein geschehen?
-Dialogue: 0,0:00:14.71,0:00:14.75,Default,,0,0,0,,Also was ist mit Ihrem wunderschönen Bein geschehen?
-Dialogue: 0,0:00:14.75,0:00:14.91,Default,,0,0,0,,Also was ist mit {\1c&HFF00&\u1}Ihrem{\r} wunderschönen Bein geschehen?
-Dialogue: 0,0:00:14.91,0:00:14.93,Default,,0,0,0,,Also was ist mit Ihrem wunderschönen Bein geschehen?
-Dialogue: 0,0:00:14.93,0:00:15.55,Default,,0,0,0,,Also was ist mit Ihrem {\1c&HFF00&\u1}wunderschönen{\r} Bein geschehen?
-Dialogue: 0,0:00:15.55,0:00:15.59,Default,,0,0,0,,Also was ist mit Ihrem wunderschönen Bein geschehen?
-Dialogue: 0,0:00:15.59,0:00:15.81,Default,,0,0,0,,Also was ist mit Ihrem wunderschönen {\1c&HFF00&\u1}Bein{\r} geschehen?
-Dialogue: 0,0:00:15.81,0:00:15.85,Default,,0,0,0,,Also was ist mit Ihrem wunderschönen Bein geschehen?
-Dialogue: 0,0:00:15.85,0:00:16.23,Default,,0,0,0,,Also was ist mit Ihrem wunderschönen Bein {\1c&HFF00&\u1}geschehen?{\r}
-Dialogue: 0,0:00:17.03,0:00:17.15,Default,,0,0,0,,{\1c&HFF00&\u1}Ein{\r} Nebenprodukt der Arschtritte, die Sie in der deutschen Filmwelt ausleihen, zweifelsohne
-Dialogue: 0,0:00:17.15,0:00:17.17,Default,,0,0,0,,Ein Nebenprodukt der Arschtritte, die Sie in der deutschen Filmwelt ausleihen, zweifelsohne
-Dialogue: 0,0:00:17.17,0:00:17.74,Default,,0,0,0,,Ein {\1c&HFF00&\u1}Nebenprodukt{\r} der Arschtritte, die Sie in der deutschen Filmwelt ausleihen, zweifelsohne
-Dialogue: 0,0:00:17.74,0:00:17.78,Default,,0,0,0,,Ein Nebenprodukt der Arschtritte, die Sie in der deutschen Filmwelt ausleihen, zweifelsohne
-Dialogue: 0,0:00:17.78,0:00:17.92,Default,,0,0,0,,Ein Nebenprodukt {\1c&HFF00&\u1}der{\r} Arschtritte, die Sie in der deutschen Filmwelt ausleihen, zweifelsohne
-Dialogue: 0,0:00:17.92,0:00:17.96,Default,,0,0,0,,Ein Nebenprodukt der Arschtritte, die Sie in der deutschen Filmwelt ausleihen, zweifelsohne
-Dialogue: 0,0:00:17.96,0:00:18.60,Default,,0,0,0,,Ein Nebenprodukt der {\1c&HFF00&\u1}Arschtritte,{\r} die Sie in der deutschen Filmwelt ausleihen, zweifelsohne
-Dialogue: 0,0:00:18.60,0:00:18.64,Default,,0,0,0,,Ein Nebenprodukt der Arschtritte, die Sie in der deutschen Filmwelt ausleihen, zweifelsohne
-Dialogue: 0,0:00:18.64,0:00:18.70,Default,,0,0,0,,Ein Nebenprodukt der Arschtritte, {\1c&HFF00&\u1}die{\r} Sie in der deutschen Filmwelt ausleihen, zweifelsohne
-Dialogue: 0,0:00:18.70,0:00:18.74,Default,,0,0,0,,Ein Nebenprodukt der Arschtritte, die Sie in der deutschen Filmwelt ausleihen, zweifelsohne
-Dialogue: 0,0:00:18.74,0:00:18.82,Default,,0,0,0,,Ein Nebenprodukt der Arschtritte, die {\1c&HFF00&\u1}Sie{\r} in der deutschen Filmwelt ausleihen, zweifelsohne
-Dialogue: 0,0:00:18.82,0:00:18.86,Default,,0,0,0,,Ein Nebenprodukt der Arschtritte, die Sie in der deutschen Filmwelt ausleihen, zweifelsohne
-Dialogue: 0,0:00:18.86,0:00:18.92,Default,,0,0,0,,Ein Nebenprodukt der Arschtritte, die Sie {\1c&HFF00&\u1}in{\r} der deutschen Filmwelt ausleihen, zweifelsohne
-Dialogue: 0,0:00:18.92,0:00:18.94,Default,,0,0,0,,Ein Nebenprodukt der Arschtritte, die Sie in der deutschen Filmwelt ausleihen, zweifelsohne
-Dialogue: 0,0:00:18.94,0:00:19.02,Default,,0,0,0,,Ein Nebenprodukt der Arschtritte, die Sie in {\1c&HFF00&\u1}der{\r} deutschen Filmwelt ausleihen, zweifelsohne
-Dialogue: 0,0:00:19.02,0:00:19.04,Default,,0,0,0,,Ein Nebenprodukt der Arschtritte, die Sie in der deutschen Filmwelt ausleihen, zweifelsohne
-Dialogue: 0,0:00:19.04,0:00:19.32,Default,,0,0,0,,Ein Nebenprodukt der Arschtritte, die Sie in der {\1c&HFF00&\u1}deutschen{\r} Filmwelt ausleihen, zweifelsohne
-Dialogue: 0,0:00:19.32,0:00:19.36,Default,,0,0,0,,Ein Nebenprodukt der Arschtritte, die Sie in der deutschen Filmwelt ausleihen, zweifelsohne
-Dialogue: 0,0:00:19.36,0:00:19.88,Default,,0,0,0,,Ein Nebenprodukt der Arschtritte, die Sie in der deutschen {\1c&HFF00&\u1}Filmwelt{\r} ausleihen, zweifelsohne
-Dialogue: 0,0:00:19.88,0:00:19.94,Default,,0,0,0,,Ein Nebenprodukt der Arschtritte, die Sie in der deutschen Filmwelt ausleihen, zweifelsohne
-Dialogue: 0,0:00:19.94,0:00:20.48,Default,,0,0,0,,Ein Nebenprodukt der Arschtritte, die Sie in der deutschen Filmwelt {\1c&HFF00&\u1}ausleihen,{\r} zweifelsohne
-Dialogue: 0,0:00:20.48,0:00:20.52,Default,,0,0,0,,Ein Nebenprodukt der Arschtritte, die Sie in der deutschen Filmwelt ausleihen, zweifelsohne
-Dialogue: 0,0:00:20.52,0:00:21.22,Default,,0,0,0,,Ein Nebenprodukt der Arschtritte, die Sie in der deutschen Filmwelt ausleihen, {\1c&HFF00&\u1}zweifelsohne{\r}
-Dialogue: 0,0:00:22.10,0:00:22.36,Default,,0,0,0,,{\1c&HFF00&\u1}Sparen{\r} Sie sich Ihre Komplimente, Sie alter Hund
-Dialogue: 0,0:00:22.36,0:00:22.38,Default,,0,0,0,,Sparen Sie sich Ihre Komplimente, Sie alter Hund
-Dialogue: 0,0:00:22.38,0:00:22.48,Default,,0,0,0,,Sparen {\1c&HFF00&\u1}Sie{\r} sich Ihre Komplimente, Sie alter Hund
-Dialogue: 0,0:00:22.48,0:00:22.50,Default,,0,0,0,,Sparen Sie sich Ihre Komplimente, Sie alter Hund
-Dialogue: 0,0:00:22.50,0:00:22.62,Default,,0,0,0,,Sparen Sie {\1c&HFF00&\u1}sich{\r} Ihre Komplimente, Sie alter Hund
-Dialogue: 0,0:00:22.62,0:00:22.64,Default,,0,0,0,,Sparen Sie sich Ihre Komplimente, Sie alter Hund
-Dialogue: 0,0:00:22.64,0:00:22.83,Default,,0,0,0,,Sparen Sie sich {\1c&HFF00&\u1}Ihre{\r} Komplimente, Sie alter Hund
-Dialogue: 0,0:00:22.83,0:00:22.87,Default,,0,0,0,,Sparen Sie sich Ihre Komplimente, Sie alter Hund
-Dialogue: 0,0:00:22.87,0:00:23.43,Default,,0,0,0,,Sparen Sie sich Ihre {\1c&HFF00&\u1}Komplimente,{\r} Sie alter Hund
-Dialogue: 0,0:00:23.43,0:00:23.45,Default,,0,0,0,,Sparen Sie sich Ihre Komplimente, Sie alter Hund
-Dialogue: 0,0:00:23.45,0:00:23.55,Default,,0,0,0,,Sparen Sie sich Ihre Komplimente, {\1c&HFF00&\u1}Sie{\r} alter Hund
-Dialogue: 0,0:00:23.55,0:00:23.59,Default,,0,0,0,,Sparen Sie sich Ihre Komplimente, Sie alter Hund
-Dialogue: 0,0:00:23.59,0:00:23.89,Default,,0,0,0,,Sparen Sie sich Ihre Komplimente, Sie {\1c&HFF00&\u1}alter{\r} Hund
-Dialogue: 0,0:00:23.89,0:00:23.93,Default,,0,0,0,,Sparen Sie sich Ihre Komplimente, Sie alter Hund
-Dialogue: 0,0:00:23.93,0:00:24.05,Default,,0,0,0,,Sparen Sie sich Ihre Komplimente, Sie alter {\1c&HFF00&\u1}Hund{\r}
-Dialogue: 0,0:00:24.57,0:00:24.65,Default,,0,0,0,,{\1c&HFF00&\u1}Ich{\r} kenne zu viele von ihren früheren Eroberungen, als dass ich in ihren Honigtopf treten könnte.
-Dialogue: 0,0:00:24.65,0:00:24.67,Default,,0,0,0,,Ich kenne zu viele von ihren früheren Eroberungen, als dass ich in ihren Honigtopf treten könnte.
-Dialogue: 0,0:00:24.67,0:00:24.97,Default,,0,0,0,,Ich {\1c&HFF00&\u1}kenne{\r} zu viele von ihren früheren Eroberungen, als dass ich in ihren Honigtopf treten könnte.
-Dialogue: 0,0:00:24.97,0:00:25.01,Default,,0,0,0,,Ich kenne zu viele von ihren früheren Eroberungen, als dass ich in ihren Honigtopf treten könnte.
-Dialogue: 0,0:00:25.01,0:00:25.19,Default,,0,0,0,,Ich kenne {\1c&HFF00&\u1}zu{\r} viele von ihren früheren Eroberungen, als dass ich in ihren Honigtopf treten könnte.
-Dialogue: 0,0:00:25.19,0:00:25.23,Default,,0,0,0,,Ich kenne zu viele von ihren früheren Eroberungen, als dass ich in ihren Honigtopf treten könnte.
-Dialogue: 0,0:00:25.23,0:00:25.49,Default,,0,0,0,,Ich kenne zu {\1c&HFF00&\u1}viele{\r} von ihren früheren Eroberungen, als dass ich in ihren Honigtopf treten könnte.
-Dialogue: 0,0:00:25.49,0:00:25.51,Default,,0,0,0,,Ich kenne zu viele von ihren früheren Eroberungen, als dass ich in ihren Honigtopf treten könnte.
-Dialogue: 0,0:00:25.51,0:00:25.65,Default,,0,0,0,,Ich kenne zu viele {\1c&HFF00&\u1}von{\r} ihren früheren Eroberungen, als dass ich in ihren Honigtopf treten könnte.
-Dialogue: 0,0:00:25.65,0:00:25.69,Default,,0,0,0,,Ich kenne zu viele von ihren früheren Eroberungen, als dass ich in ihren Honigtopf treten könnte.
-Dialogue: 0,0:00:25.69,0:00:25.95,Default,,0,0,0,,Ich kenne zu viele von {\1c&HFF00&\u1}ihren{\r} früheren Eroberungen, als dass ich in ihren Honigtopf treten könnte.
-Dialogue: 0,0:00:25.95,0:00:25.99,Default,,0,0,0,,Ich kenne zu viele von ihren früheren Eroberungen, als dass ich in ihren Honigtopf treten könnte.
-Dialogue: 0,0:00:25.99,0:00:26.41,Default,,0,0,0,,Ich kenne zu viele von ihren {\1c&HFF00&\u1}früheren{\r} Eroberungen, als dass ich in ihren Honigtopf treten könnte.
-Dialogue: 0,0:00:26.41,0:00:26.45,Default,,0,0,0,,Ich kenne zu viele von ihren früheren Eroberungen, als dass ich in ihren Honigtopf treten könnte.
-Dialogue: 0,0:00:26.45,0:00:27.01,Default,,0,0,0,,Ich kenne zu viele von ihren früheren {\1c&HFF00&\u1}Eroberungen,{\r} als dass ich in ihren Honigtopf treten könnte.
-Dialogue: 0,0:00:27.01,0:00:27.06,Default,,0,0,0,,Ich kenne zu viele von ihren früheren Eroberungen, als dass ich in ihren Honigtopf treten könnte.
-Dialogue: 0,0:00:27.06,0:00:27.24,Default,,0,0,0,,Ich kenne zu viele von ihren früheren Eroberungen, {\1c&HFF00&\u1}als{\r} dass ich in ihren Honigtopf treten könnte.
-Dialogue: 0,0:00:27.24,0:00:27.26,Default,,0,0,0,,Ich kenne zu viele von ihren früheren Eroberungen, als dass ich in ihren Honigtopf treten könnte.
-Dialogue: 0,0:00:27.26,0:00:27.46,Default,,0,0,0,,Ich kenne zu viele von ihren früheren Eroberungen, als {\1c&HFF00&\u1}dass{\r} ich in ihren Honigtopf treten könnte.
-Dialogue: 0,0:00:27.46,0:00:27.50,Default,,0,0,0,,Ich kenne zu viele von ihren früheren Eroberungen, als dass ich in ihren Honigtopf treten könnte.
-Dialogue: 0,0:00:27.50,0:00:27.60,Default,,0,0,0,,Ich kenne zu viele von ihren früheren Eroberungen, als dass {\1c&HFF00&\u1}ich{\r} in ihren Honigtopf treten könnte.
-Dialogue: 0,0:00:27.60,0:00:27.64,Default,,0,0,0,,Ich kenne zu viele von ihren früheren Eroberungen, als dass ich in ihren Honigtopf treten könnte.
-Dialogue: 0,0:00:27.64,0:00:27.74,Default,,0,0,0,,Ich kenne zu viele von ihren früheren Eroberungen, als dass ich {\1c&HFF00&\u1}in{\r} ihren Honigtopf treten könnte.
-Dialogue: 0,0:00:27.74,0:00:27.78,Default,,0,0,0,,Ich kenne zu viele von ihren früheren Eroberungen, als dass ich in ihren Honigtopf treten könnte.
-Dialogue: 0,0:00:27.78,0:00:28.00,Default,,0,0,0,,Ich kenne zu viele von ihren früheren Eroberungen, als dass ich in {\1c&HFF00&\u1}ihren{\r} Honigtopf treten könnte.
-Dialogue: 0,0:00:28.00,0:00:28.04,Default,,0,0,0,,Ich kenne zu viele von ihren früheren Eroberungen, als dass ich in ihren Honigtopf treten könnte.
-Dialogue: 0,0:00:28.04,0:00:28.64,Default,,0,0,0,,Ich kenne zu viele von ihren früheren Eroberungen, als dass ich in ihren {\1c&HFF00&\u1}Honigtopf{\r} treten könnte.
-Dialogue: 0,0:00:28.64,0:00:28.68,Default,,0,0,0,,Ich kenne zu viele von ihren früheren Eroberungen, als dass ich in ihren Honigtopf treten könnte.
-Dialogue: 0,0:00:28.68,0:00:28.96,Default,,0,0,0,,Ich kenne zu viele von ihren früheren Eroberungen, als dass ich in ihren Honigtopf {\1c&HFF00&\u1}treten{\r} könnte.
-Dialogue: 0,0:00:28.96,0:00:29.00,Default,,0,0,0,,Ich kenne zu viele von ihren früheren Eroberungen, als dass ich in ihren Honigtopf treten könnte.
-Dialogue: 0,0:00:29.00,0:00:29.24,Default,,0,0,0,,Ich kenne zu viele von ihren früheren Eroberungen, als dass ich in ihren Honigtopf treten {\1c&HFF00&\u1}könnte.{\r}
-Dialogue: 0,0:00:29.24,0:00:29.30,Default,,0,0,0,,{\1c&HFF00&\u1}Na{\r} im Ernst, was ist passiert?
-Dialogue: 0,0:00:29.30,0:00:29.32,Default,,0,0,0,,Na im Ernst, was ist passiert?
-Dialogue: 0,0:00:29.32,0:00:29.36,Default,,0,0,0,,Na {\1c&HFF00&\u1}im{\r} Ernst, was ist passiert?
-Dialogue: 0,0:00:29.36,0:00:29.38,Default,,0,0,0,,Na im Ernst, was ist passiert?
-Dialogue: 0,0:00:29.38,0:00:29.48,Default,,0,0,0,,Na im {\1c&HFF00&\u1}Ernst,{\r} was ist passiert?
-Dialogue: 0,0:00:29.48,0:00:29.50,Default,,0,0,0,,Na im Ernst, was ist passiert?
-Dialogue: 0,0:00:29.50,0:00:29.56,Default,,0,0,0,,Na im Ernst, {\1c&HFF00&\u1}was{\r} ist passiert?
-Dialogue: 0,0:00:29.56,0:00:29.58,Default,,0,0,0,,Na im Ernst, was ist passiert?
-Dialogue: 0,0:00:29.58,0:00:29.64,Default,,0,0,0,,Na im Ernst, was {\1c&HFF00&\u1}ist{\r} passiert?
-Dialogue: 0,0:00:29.64,0:00:29.66,Default,,0,0,0,,Na im Ernst, was ist passiert?
-Dialogue: 0,0:00:29.66,0:00:29.82,Default,,0,0,0,,Na im Ernst, was ist {\1c&HFF00&\u1}passiert?{\r}
-Dialogue: 0,0:00:30.78,0:00:32.27,Default,,0,0,0,,{\1c&HFF00&\u1}Tja,{\r} ich habe mich, dummerweise muss ich eingestehen, im Bergsteigen versucht.
-Dialogue: 0,0:00:32.27,0:00:32.33,Default,,0,0,0,,Tja, ich habe mich, dummerweise muss ich eingestehen, im Bergsteigen versucht.
-Dialogue: 0,0:00:32.33,0:00:32.49,Default,,0,0,0,,Tja, {\1c&HFF00&\u1}ich{\r} habe mich, dummerweise muss ich eingestehen, im Bergsteigen versucht.
-Dialogue: 0,0:00:32.49,0:00:32.53,Default,,0,0,0,,Tja, ich habe mich, dummerweise muss ich eingestehen, im Bergsteigen versucht.
-Dialogue: 0,0:00:32.53,0:00:32.79,Default,,0,0,0,,Tja, ich {\1c&HFF00&\u1}habe{\r} mich, dummerweise muss ich eingestehen, im Bergsteigen versucht.
-Dialogue: 0,0:00:32.79,0:00:32.83,Default,,0,0,0,,Tja, ich habe mich, dummerweise muss ich eingestehen, im Bergsteigen versucht.
-Dialogue: 0,0:00:32.83,0:00:33.83,Default,,0,0,0,,Tja, ich habe {\1c&HFF00&\u1}mich,{\r} dummerweise muss ich eingestehen, im Bergsteigen versucht.
-Dialogue: 0,0:00:33.83,0:00:33.85,Default,,0,0,0,,Tja, ich habe mich, dummerweise muss ich eingestehen, im Bergsteigen versucht.
-Dialogue: 0,0:00:33.85,0:00:34.57,Default,,0,0,0,,Tja, ich habe mich, {\1c&HFF00&\u1}dummerweise{\r} muss ich eingestehen, im Bergsteigen versucht.
-Dialogue: 0,0:00:34.57,0:00:34.59,Default,,0,0,0,,Tja, ich habe mich, dummerweise muss ich eingestehen, im Bergsteigen versucht.
-Dialogue: 0,0:00:34.59,0:00:34.73,Default,,0,0,0,,Tja, ich habe mich, dummerweise {\1c&HFF00&\u1}muss{\r} ich eingestehen, im Bergsteigen versucht.
-Dialogue: 0,0:00:34.73,0:00:34.77,Default,,0,0,0,,Tja, ich habe mich, dummerweise muss ich eingestehen, im Bergsteigen versucht.
-Dialogue: 0,0:00:34.77,0:00:34.89,Default,,0,0,0,,Tja, ich habe mich, dummerweise muss {\1c&HFF00&\u1}ich{\r} eingestehen, im Bergsteigen versucht.
-Dialogue: 0,0:00:34.89,0:00:34.93,Default,,0,0,0,,Tja, ich habe mich, dummerweise muss ich eingestehen, im Bergsteigen versucht.
-Dialogue: 0,0:00:34.93,0:00:36.83,Default,,0,0,0,,Tja, ich habe mich, dummerweise muss ich {\1c&HFF00&\u1}eingestehen,{\r} im Bergsteigen versucht.
-Dialogue: 0,0:00:36.83,0:00:36.87,Default,,0,0,0,,Tja, ich habe mich, dummerweise muss ich eingestehen, im Bergsteigen versucht.
-Dialogue: 0,0:00:36.87,0:00:36.99,Default,,0,0,0,,Tja, ich habe mich, dummerweise muss ich eingestehen, {\1c&HFF00&\u1}im{\r} Bergsteigen versucht.
-Dialogue: 0,0:00:36.99,0:00:37.03,Default,,0,0,0,,Tja, ich habe mich, dummerweise muss ich eingestehen, im Bergsteigen versucht.
-Dialogue: 0,0:00:37.03,0:00:37.76,Default,,0,0,0,,Tja, ich habe mich, dummerweise muss ich eingestehen, im {\1c&HFF00&\u1}Bergsteigen{\r} versucht.
-Dialogue: 0,0:00:37.76,0:00:37.78,Default,,0,0,0,,Tja, ich habe mich, dummerweise muss ich eingestehen, im Bergsteigen versucht.
-Dialogue: 0,0:00:37.78,0:00:38.22,Default,,0,0,0,,Tja, ich habe mich, dummerweise muss ich eingestehen, im Bergsteigen {\1c&HFF00&\u1}versucht.{\r}
-Dialogue: 0,0:00:41.23,0:00:41.85,Default,,0,0,0,,{\1c&HFF00&\u1}Bergsteigen?{\r} Dabei haben sie ihr Bein verletzt beim Bergsteigen?
-Dialogue: 0,0:00:41.85,0:00:41.87,Default,,0,0,0,,Bergsteigen? Dabei haben sie ihr Bein verletzt beim Bergsteigen?
-Dialogue: 0,0:00:41.87,0:00:42.13,Default,,0,0,0,,Bergsteigen? {\1c&HFF00&\u1}Dabei{\r} haben sie ihr Bein verletzt beim Bergsteigen?
-Dialogue: 0,0:00:42.13,0:00:42.15,Default,,0,0,0,,Bergsteigen? Dabei haben sie ihr Bein verletzt beim Bergsteigen?
-Dialogue: 0,0:00:42.15,0:00:42.33,Default,,0,0,0,,Bergsteigen? Dabei {\1c&HFF00&\u1}haben{\r} sie ihr Bein verletzt beim Bergsteigen?
-Dialogue: 0,0:00:42.33,0:00:42.37,Default,,0,0,0,,Bergsteigen? Dabei haben sie ihr Bein verletzt beim Bergsteigen?
-Dialogue: 0,0:00:42.37,0:00:42.51,Default,,0,0,0,,Bergsteigen? Dabei haben {\1c&HFF00&\u1}sie{\r} ihr Bein verletzt beim Bergsteigen?
-Dialogue: 0,0:00:42.51,0:00:42.55,Default,,0,0,0,,Bergsteigen? Dabei haben sie ihr Bein verletzt beim Bergsteigen?
-Dialogue: 0,0:00:42.55,0:00:42.61,Default,,0,0,0,,Bergsteigen? Dabei haben sie {\1c&HFF00&\u1}ihr{\r} Bein verletzt beim Bergsteigen?
-Dialogue: 0,0:00:42.61,0:00:42.63,Default,,0,0,0,,Bergsteigen? Dabei haben sie ihr Bein verletzt beim Bergsteigen?
-Dialogue: 0,0:00:42.63,0:00:42.77,Default,,0,0,0,,Bergsteigen? Dabei haben sie ihr {\1c&HFF00&\u1}Bein{\r} verletzt beim Bergsteigen?
-Dialogue: 0,0:00:42.77,0:00:42.81,Default,,0,0,0,,Bergsteigen? Dabei haben sie ihr Bein verletzt beim Bergsteigen?
-Dialogue: 0,0:00:42.81,0:00:43.23,Default,,0,0,0,,Bergsteigen? Dabei haben sie ihr Bein {\1c&HFF00&\u1}verletzt{\r} beim Bergsteigen?
-Dialogue: 0,0:00:43.23,0:00:43.27,Default,,0,0,0,,Bergsteigen? Dabei haben sie ihr Bein verletzt beim Bergsteigen?
-Dialogue: 0,0:00:43.27,0:00:43.39,Default,,0,0,0,,Bergsteigen? Dabei haben sie ihr Bein verletzt {\1c&HFF00&\u1}beim{\r} Bergsteigen?
-Dialogue: 0,0:00:43.39,0:00:43.45,Default,,0,0,0,,Bergsteigen? Dabei haben sie ihr Bein verletzt beim Bergsteigen?
-Dialogue: 0,0:00:43.45,0:00:44.13,Default,,0,0,0,,Bergsteigen? Dabei haben sie ihr Bein verletzt beim {\1c&HFF00&\u1}Bergsteigen?{\r}
-Dialogue: 0,0:00:44.54,0:00:44.62,Default,,0,0,0,,{\1c&HFF00&\u1}Ob{\r} sie es glauben oder nicht.
-Dialogue: 0,0:00:44.62,0:00:44.66,Default,,0,0,0,,Ob sie es glauben oder nicht.
-Dialogue: 0,0:00:44.66,0:00:44.72,Default,,0,0,0,,Ob {\1c&HFF00&\u1}sie{\r} es glauben oder nicht.
-Dialogue: 0,0:00:44.72,0:00:44.74,Default,,0,0,0,,Ob sie es glauben oder nicht.
-Dialogue: 0,0:00:44.74,0:00:44.80,Default,,0,0,0,,Ob sie {\1c&HFF00&\u1}es{\r} glauben oder nicht.
-Dialogue: 0,0:00:44.80,0:00:44.82,Default,,0,0,0,,Ob sie es glauben oder nicht.
-Dialogue: 0,0:00:44.82,0:00:45.14,Default,,0,0,0,,Ob sie es {\1c&HFF00&\u1}glauben{\r} oder nicht.
-Dialogue: 0,0:00:45.14,0:00:45.16,Default,,0,0,0,,Ob sie es glauben oder nicht.
-Dialogue: 0,0:00:45.16,0:00:45.46,Default,,0,0,0,,Ob sie es glauben {\1c&HFF00&\u1}oder{\r} nicht.
-Dialogue: 0,0:00:45.46,0:00:45.50,Default,,0,0,0,,Ob sie es glauben oder nicht.
-Dialogue: 0,0:00:45.50,0:00:45.66,Default,,0,0,0,,Ob sie es glauben oder {\1c&HFF00&\u1}nicht.{\r}

examples/whisperx/sample_de_01.wav.srt

@@ -1,44 +0,0 @@
-1
-00:00:00,563 --> 00:00:01,869
-Weinlein von Hammersmann
-
-2
-00:00:06,187 --> 00:00:08,013
-Oberst Lande, es ist lange her
-
-3
-00:00:08,013 --> 00:00:08,534
-Schneide ich wie eh und je
-
-4
-00:00:13,987 --> 00:00:16,234
-Also was ist mit Ihrem wunderschönen Bein geschehen?
-
-5
-00:00:17,035 --> 00:00:21,218
-Ein Nebenprodukt der Arschtritte, die Sie in der deutschen Filmwelt ausleihen, zweifelsohne
-
-6
-00:00:22,102 --> 00:00:24,051
-Sparen Sie sich Ihre Komplimente, Sie alter Hund
-
-7
-00:00:24,572 --> 00:00:29,238
-Ich kenne zu viele von ihren früheren Eroberungen, als dass ich in ihren Honigtopf treten könnte.
-
-8
-00:00:29,238 --> 00:00:29,821
-Na im Ernst, was ist passiert?
-
-9
-00:00:30,783 --> 00:00:38,217
-Tja, ich habe mich, dummerweise muss ich eingestehen, im Bergsteigen versucht.
-
-10
-00:00:41,226 --> 00:00:44,135
-Bergsteigen? Dabei haben sie ihr Bein verletzt beim Bergsteigen?
-
-11
-00:00:44,535 --> 00:00:45,657
-Ob sie es glauben oder nicht.
-

examples/whisperx/sample_fr_01.wav.ass

@@ -1,199 +0,0 @@
-[Script Info]
-ScriptType: v4.00+
-PlayResX: 384
-PlayResY: 288
-ScaledBorderAndShadow: yes
-
-[V4+ Styles]
-Format: Name, Fontname, Fontsize, PrimaryColour, SecondaryColour, OutlineColour, BackColour, Bold, Italic, Underline, StrikeOut, ScaleX, ScaleY, Spacing, Angle, BorderStyle, Outline, Shadow, Alignment, MarginL, MarginR, MarginV, Encoding
-Style: Default,Arial,24,&Hffffff,&Hffffff,&H0,&H0,0,0,0,0,100,100,0,0,1,1,0,2,10,10,10,0
-
-[Events]
-Format: Layer, Start, End, Style, Name, MarginL, MarginR, MarginV, Effect, Text
-
-Dialogue: 0,0:00:0.77,0:00:1.07,Default,,0,0,0,,{\1c&HFF00&\u1}Lâchez,{\r} c'est bon.
-Dialogue: 0,0:00:1.07,0:00:1.11,Default,,0,0,0,,Lâchez, c'est bon.
-Dialogue: 0,0:00:1.11,0:00:1.21,Default,,0,0,0,,Lâchez, {\1c&HFF00&\u1}c'est{\r} bon.
-Dialogue: 0,0:00:1.21,0:00:1.23,Default,,0,0,0,,Lâchez, c'est bon.
-Dialogue: 0,0:00:1.23,0:00:1.31,Default,,0,0,0,,Lâchez, c'est {\1c&HFF00&\u1}bon.{\r}
-Dialogue: 0,0:00:1.31,0:00:1.81,Default,,0,0,0,,{\1c&HFF00&\u1}Ça{\r} va?
-Dialogue: 0,0:00:1.81,0:00:1.83,Default,,0,0,0,,Ça va?
-Dialogue: 0,0:00:1.83,0:00:1.89,Default,,0,0,0,,Ça {\1c&HFF00&\u1}va?{\r}
-Dialogue: 0,0:00:1.91,0:00:1.97,Default,,0,0,0,,{\1c&HFF00&\u1}Oui.{\r}
-Dialogue: 0,0:00:1.97,0:00:3.27,Default,,0,0,0,,{\1c&HFF00&\u1}Merci{\r} beaucoup.
-Dialogue: 0,0:00:3.27,0:00:3.29,Default,,0,0,0,,Merci beaucoup.
-Dialogue: 0,0:00:3.29,0:00:3.49,Default,,0,0,0,,Merci {\1c&HFF00&\u1}beaucoup.{\r}
-Dialogue: 0,0:00:4.36,0:00:4.58,Default,,0,0,0,,{\1c&HFF00&\u1}Chèque{\r} ou espèce?
-Dialogue: 0,0:00:4.58,0:00:4.64,Default,,0,0,0,,Chèque ou espèce?
-Dialogue: 0,0:00:4.64,0:00:4.72,Default,,0,0,0,,Chèque {\1c&HFF00&\u1}ou{\r} espèce?
-Dialogue: 0,0:00:4.72,0:00:4.78,Default,,0,0,0,,Chèque ou espèce?
-Dialogue: 0,0:00:4.78,0:00:5.04,Default,,0,0,0,,Chèque ou {\1c&HFF00&\u1}espèce?{\r}
-Dialogue: 0,0:00:6.54,0:00:6.70,Default,,0,0,0,,{\1c&HFF00&\u1}J'ai{\r} laissé un chèque sur la commode, il est signé.
-Dialogue: 0,0:00:6.70,0:00:6.74,Default,,0,0,0,,J'ai laissé un chèque sur la commode, il est signé.
-Dialogue: 0,0:00:6.74,0:00:6.99,Default,,0,0,0,,J'ai {\1c&HFF00&\u1}laissé{\r} un chèque sur la commode, il est signé.
-Dialogue: 0,0:00:6.99,0:00:7.03,Default,,0,0,0,,J'ai laissé un chèque sur la commode, il est signé.
-Dialogue: 0,0:00:7.03,0:00:7.09,Default,,0,0,0,,J'ai laissé {\1c&HFF00&\u1}un{\r} chèque sur la commode, il est signé.
-Dialogue: 0,0:00:7.09,0:00:7.13,Default,,0,0,0,,J'ai laissé un chèque sur la commode, il est signé.
-Dialogue: 0,0:00:7.13,0:00:7.33,Default,,0,0,0,,J'ai laissé un {\1c&HFF00&\u1}chèque{\r} sur la commode, il est signé.
-Dialogue: 0,0:00:7.33,0:00:7.35,Default,,0,0,0,,J'ai laissé un chèque sur la commode, il est signé.
-Dialogue: 0,0:00:7.35,0:00:7.49,Default,,0,0,0,,J'ai laissé un chèque {\1c&HFF00&\u1}sur{\r} la commode, il est signé.
-Dialogue: 0,0:00:7.49,0:00:7.51,Default,,0,0,0,,J'ai laissé un chèque sur la commode, il est signé.
-Dialogue: 0,0:00:7.51,0:00:7.59,Default,,0,0,0,,J'ai laissé un chèque sur {\1c&HFF00&\u1}la{\r} commode, il est signé.
-Dialogue: 0,0:00:7.59,0:00:7.63,Default,,0,0,0,,J'ai laissé un chèque sur la commode, il est signé.
-Dialogue: 0,0:00:7.63,0:00:7.91,Default,,0,0,0,,J'ai laissé un chèque sur la {\1c&HFF00&\u1}commode,{\r} il est signé.
-Dialogue: 0,0:00:7.91,0:00:7.99,Default,,0,0,0,,J'ai laissé un chèque sur la commode, il est signé.
-Dialogue: 0,0:00:7.99,0:00:8.03,Default,,0,0,0,,J'ai laissé un chèque sur la commode, {\1c&HFF00&\u1}il{\r} est signé.
-Dialogue: 0,0:00:8.03,0:00:8.09,Default,,0,0,0,,J'ai laissé un chèque sur la commode, il est signé.
-Dialogue: 0,0:00:8.09,0:00:8.19,Default,,0,0,0,,J'ai laissé un chèque sur la commode, il {\1c&HFF00&\u1}est{\r} signé.
-Dialogue: 0,0:00:8.19,0:00:8.21,Default,,0,0,0,,J'ai laissé un chèque sur la commode, il est signé.
-Dialogue: 0,0:00:8.21,0:00:8.39,Default,,0,0,0,,J'ai laissé un chèque sur la commode, il est {\1c&HFF00&\u1}signé.{\r}
-Dialogue: 0,0:00:8.39,0:00:8.81,Default,,0,0,0,,{\1c&HFF00&\u1}Vous{\r} le remplirez.
-Dialogue: 0,0:00:8.81,0:00:8.83,Default,,0,0,0,,Vous le remplirez.
-Dialogue: 0,0:00:8.83,0:00:8.95,Default,,0,0,0,,Vous {\1c&HFF00&\u1}le{\r} remplirez.
-Dialogue: 0,0:00:8.95,0:00:8.97,Default,,0,0,0,,Vous le remplirez.
-Dialogue: 0,0:00:8.97,0:00:9.26,Default,,0,0,0,,Vous le {\1c&HFF00&\u1}remplirez.{\r}
-Dialogue: 0,0:00:9.28,0:00:9.34,Default,,0,0,0,,{\1c&HFF00&\u1}OK.{\r}
-Dialogue: 0,0:00:9.36,0:00:9.40,Default,,0,0,0,,{\1c&HFF00&\u1}Oh!{\r}
-Dialogue: 0,0:00:12.41,0:00:12.51,Default,,0,0,0,,{\1c&HFF00&\u1}Ouh{\r} là!
-Dialogue: 0,0:00:12.51,0:00:12.53,Default,,0,0,0,,Ouh là!
-Dialogue: 0,0:00:12.53,0:00:12.59,Default,,0,0,0,,Ouh {\1c&HFF00&\u1}là!{\r}
-Dialogue: 0,0:00:12.59,0:00:12.73,Default,,0,0,0,,{\1c&HFF00&\u1}Venez.{\r}
-Dialogue: 0,0:00:14.45,0:00:14.63,Default,,0,0,0,,{\1c&HFF00&\u1}Merci.{\r}
-Dialogue: 0,0:00:14.65,0:00:14.76,Default,,0,0,0,,{\1c&HFF00&\u1}Ah!{\r}
-Dialogue: 0,0:00:15.64,0:00:16.51,Default,,0,0,0,,{\1c&HFF00&\u1}C'est{\r} qui?
-Dialogue: 0,0:00:16.51,0:00:16.53,Default,,0,0,0,,C'est qui?
-Dialogue: 0,0:00:16.53,0:00:16.63,Default,,0,0,0,,C'est {\1c&HFF00&\u1}qui?{\r}
-Dialogue: 0,0:00:20.00,0:00:22.85,Default,,0,0,0,,{\1c&HFF00&\u1}C'est{\r} pas vrai, qu'est-ce qu'il fout ici, ce con?
-Dialogue: 0,0:00:22.85,0:00:22.87,Default,,0,0,0,,C'est pas vrai, qu'est-ce qu'il fout ici, ce con?
-Dialogue: 0,0:00:22.87,0:00:22.93,Default,,0,0,0,,C'est {\1c&HFF00&\u1}pas{\r} vrai, qu'est-ce qu'il fout ici, ce con?
-Dialogue: 0,0:00:22.93,0:00:22.95,Default,,0,0,0,,C'est pas vrai, qu'est-ce qu'il fout ici, ce con?
-Dialogue: 0,0:00:22.95,0:00:23.15,Default,,0,0,0,,C'est pas {\1c&HFF00&\u1}vrai,{\r} qu'est-ce qu'il fout ici, ce con?
-Dialogue: 0,0:00:23.15,0:00:23.17,Default,,0,0,0,,C'est pas vrai, qu'est-ce qu'il fout ici, ce con?
-Dialogue: 0,0:00:23.17,0:00:23.35,Default,,0,0,0,,C'est pas vrai, {\1c&HFF00&\u1}qu'est-ce{\r} qu'il fout ici, ce con?
-Dialogue: 0,0:00:23.35,0:00:23.37,Default,,0,0,0,,C'est pas vrai, qu'est-ce qu'il fout ici, ce con?
-Dialogue: 0,0:00:23.37,0:00:23.51,Default,,0,0,0,,C'est pas vrai, qu'est-ce {\1c&HFF00&\u1}qu'il{\r} fout ici, ce con?
-Dialogue: 0,0:00:23.51,0:00:23.53,Default,,0,0,0,,C'est pas vrai, qu'est-ce qu'il fout ici, ce con?
-Dialogue: 0,0:00:23.53,0:00:23.67,Default,,0,0,0,,C'est pas vrai, qu'est-ce qu'il {\1c&HFF00&\u1}fout{\r} ici, ce con?
-Dialogue: 0,0:00:23.67,0:00:23.73,Default,,0,0,0,,C'est pas vrai, qu'est-ce qu'il fout ici, ce con?
-Dialogue: 0,0:00:23.73,0:00:23.95,Default,,0,0,0,,C'est pas vrai, qu'est-ce qu'il fout {\1c&HFF00&\u1}ici,{\r} ce con?
-Dialogue: 0,0:00:23.95,0:00:23.99,Default,,0,0,0,,C'est pas vrai, qu'est-ce qu'il fout ici, ce con?
-Dialogue: 0,0:00:23.99,0:00:24.11,Default,,0,0,0,,C'est pas vrai, qu'est-ce qu'il fout ici, {\1c&HFF00&\u1}ce{\r} con?
-Dialogue: 0,0:00:24.11,0:00:24.15,Default,,0,0,0,,C'est pas vrai, qu'est-ce qu'il fout ici, ce con?
-Dialogue: 0,0:00:24.15,0:00:24.23,Default,,0,0,0,,C'est pas vrai, qu'est-ce qu'il fout ici, ce {\1c&HFF00&\u1}con?{\r}
-Dialogue: 0,0:00:24.51,0:00:24.96,Default,,0,0,0,,{\1c&HFF00&\u1}Excusez-moi,{\r} mais je crois que j'ai oublié mon sac chez vous.
-Dialogue: 0,0:00:24.96,0:00:24.98,Default,,0,0,0,,Excusez-moi, mais je crois que j'ai oublié mon sac chez vous.
-Dialogue: 0,0:00:24.98,0:00:25.06,Default,,0,0,0,,Excusez-moi, {\1c&HFF00&\u1}mais{\r} je crois que j'ai oublié mon sac chez vous.
-Dialogue: 0,0:00:25.06,0:00:25.08,Default,,0,0,0,,Excusez-moi, mais je crois que j'ai oublié mon sac chez vous.
-Dialogue: 0,0:00:25.08,0:00:25.12,Default,,0,0,0,,Excusez-moi, mais {\1c&HFF00&\u1}je{\r} crois que j'ai oublié mon sac chez vous.
-Dialogue: 0,0:00:25.12,0:00:25.14,Default,,0,0,0,,Excusez-moi, mais je crois que j'ai oublié mon sac chez vous.
-Dialogue: 0,0:00:25.14,0:00:25.26,Default,,0,0,0,,Excusez-moi, mais je {\1c&HFF00&\u1}crois{\r} que j'ai oublié mon sac chez vous.
-Dialogue: 0,0:00:25.26,0:00:25.28,Default,,0,0,0,,Excusez-moi, mais je crois que j'ai oublié mon sac chez vous.
-Dialogue: 0,0:00:25.28,0:00:25.34,Default,,0,0,0,,Excusez-moi, mais je crois {\1c&HFF00&\u1}que{\r} j'ai oublié mon sac chez vous.
-Dialogue: 0,0:00:25.34,0:00:25.36,Default,,0,0,0,,Excusez-moi, mais je crois que j'ai oublié mon sac chez vous.
-Dialogue: 0,0:00:25.36,0:00:25.42,Default,,0,0,0,,Excusez-moi, mais je crois que {\1c&HFF00&\u1}j'ai{\r} oublié mon sac chez vous.
-Dialogue: 0,0:00:25.42,0:00:25.44,Default,,0,0,0,,Excusez-moi, mais je crois que j'ai oublié mon sac chez vous.
-Dialogue: 0,0:00:25.44,0:00:25.60,Default,,0,0,0,,Excusez-moi, mais je crois que j'ai {\1c&HFF00&\u1}oublié{\r} mon sac chez vous.
-Dialogue: 0,0:00:25.60,0:00:25.62,Default,,0,0,0,,Excusez-moi, mais je crois que j'ai oublié mon sac chez vous.
-Dialogue: 0,0:00:25.62,0:00:25.76,Default,,0,0,0,,Excusez-moi, mais je crois que j'ai oublié {\1c&HFF00&\u1}mon{\r} sac chez vous.
-Dialogue: 0,0:00:25.76,0:00:25.78,Default,,0,0,0,,Excusez-moi, mais je crois que j'ai oublié mon sac chez vous.
-Dialogue: 0,0:00:25.78,0:00:25.94,Default,,0,0,0,,Excusez-moi, mais je crois que j'ai oublié mon {\1c&HFF00&\u1}sac{\r} chez vous.
-Dialogue: 0,0:00:25.94,0:00:25.96,Default,,0,0,0,,Excusez-moi, mais je crois que j'ai oublié mon sac chez vous.
-Dialogue: 0,0:00:25.96,0:00:26.04,Default,,0,0,0,,Excusez-moi, mais je crois que j'ai oublié mon sac {\1c&HFF00&\u1}chez{\r} vous.
-Dialogue: 0,0:00:26.04,0:00:26.06,Default,,0,0,0,,Excusez-moi, mais je crois que j'ai oublié mon sac chez vous.
-Dialogue: 0,0:00:26.06,0:00:26.18,Default,,0,0,0,,Excusez-moi, mais je crois que j'ai oublié mon sac chez {\1c&HFF00&\u1}vous.{\r}
-Dialogue: 0,0:00:26.18,0:00:26.30,Default,,0,0,0,,{\1c&HFF00&\u1}Ça{\r} va?
-Dialogue: 0,0:00:26.30,0:00:26.32,Default,,0,0,0,,Ça va?
-Dialogue: 0,0:00:26.32,0:00:26.36,Default,,0,0,0,,Ça {\1c&HFF00&\u1}va?{\r}
-Dialogue: 0,0:00:31.04,0:00:31.24,Default,,0,0,0,,{\1c&HFF00&\u1}Attendez.{\r}
-Dialogue: 0,0:00:36.81,0:00:36.97,Default,,0,0,0,,{\1c&HFF00&\u1}Tout{\r} à l'heure, là, c'était...
-Dialogue: 0,0:00:36.97,0:00:37.01,Default,,0,0,0,,Tout à l'heure, là, c'était...
-Dialogue: 0,0:00:37.01,0:00:37.05,Default,,0,0,0,,Tout {\1c&HFF00&\u1}à{\r} l'heure, là, c'était...
-Dialogue: 0,0:00:37.05,0:00:37.09,Default,,0,0,0,,Tout à l'heure, là, c'était...
-Dialogue: 0,0:00:37.09,0:00:37.35,Default,,0,0,0,,Tout à {\1c&HFF00&\u1}l'heure,{\r} là, c'était...
-Dialogue: 0,0:00:37.35,0:00:37.39,Default,,0,0,0,,Tout à l'heure, là, c'était...
-Dialogue: 0,0:00:37.39,0:00:37.98,Default,,0,0,0,,Tout à l'heure, {\1c&HFF00&\u1}là,{\r} c'était...
-Dialogue: 0,0:00:37.98,0:00:38.02,Default,,0,0,0,,Tout à l'heure, là, c'était...
-Dialogue: 0,0:00:38.02,0:00:38.28,Default,,0,0,0,,Tout à l'heure, là, {\1c&HFF00&\u1}c'était...{\r}
-Dialogue: 0,0:00:38.28,0:00:38.36,Default,,0,0,0,,{\1c&HFF00&\u1}Vous?{\r}
-Dialogue: 0,0:00:39.12,0:00:39.24,Default,,0,0,0,,{\1c&HFF00&\u1}Vous?{\r} Pas...
-Dialogue: 0,0:00:39.24,0:00:39.30,Default,,0,0,0,,Vous? Pas...
-Dialogue: 0,0:00:39.30,0:00:39.42,Default,,0,0,0,,Vous? {\1c&HFF00&\u1}Pas...{\r}
-Dialogue: 0,0:00:39.42,0:00:39.53,Default,,0,0,0,,{\1c&HFF00&\u1}Pas{\r} lui? Vous?
-Dialogue: 0,0:00:39.53,0:00:39.55,Default,,0,0,0,,Pas lui? Vous?
-Dialogue: 0,0:00:39.55,0:00:39.61,Default,,0,0,0,,Pas {\1c&HFF00&\u1}lui?{\r} Vous?
-Dialogue: 0,0:00:39.61,0:00:39.63,Default,,0,0,0,,Pas lui? Vous?
-Dialogue: 0,0:00:39.63,0:00:39.71,Default,,0,0,0,,Pas lui? {\1c&HFF00&\u1}Vous?{\r}
-Dialogue: 0,0:00:44.19,0:00:44.35,Default,,0,0,0,,{\1c&HFF00&\u1}Vous{\r} avez tout à fait raison, M. Xenakis.
-Dialogue: 0,0:00:44.35,0:00:44.39,Default,,0,0,0,,Vous avez tout à fait raison, M. Xenakis.
-Dialogue: 0,0:00:44.39,0:00:44.62,Default,,0,0,0,,Vous {\1c&HFF00&\u1}avez{\r} tout à fait raison, M. Xenakis.
-Dialogue: 0,0:00:44.62,0:00:44.64,Default,,0,0,0,,Vous avez tout à fait raison, M. Xenakis.
-Dialogue: 0,0:00:44.64,0:00:44.80,Default,,0,0,0,,Vous avez {\1c&HFF00&\u1}tout{\r} à fait raison, M. Xenakis.
-Dialogue: 0,0:00:44.80,0:00:44.82,Default,,0,0,0,,Vous avez tout à fait raison, M. Xenakis.
-Dialogue: 0,0:00:44.82,0:00:44.84,Default,,0,0,0,,Vous avez tout {\1c&HFF00&\u1}à{\r} fait raison, M. Xenakis.
-Dialogue: 0,0:00:44.84,0:00:44.90,Default,,0,0,0,,Vous avez tout à fait raison, M. Xenakis.
-Dialogue: 0,0:00:44.90,0:00:45.04,Default,,0,0,0,,Vous avez tout à {\1c&HFF00&\u1}fait{\r} raison, M. Xenakis.
-Dialogue: 0,0:00:45.04,0:00:45.08,Default,,0,0,0,,Vous avez tout à fait raison, M. Xenakis.
-Dialogue: 0,0:00:45.08,0:00:45.38,Default,,0,0,0,,Vous avez tout à fait {\1c&HFF00&\u1}raison,{\r} M. Xenakis.
-Dialogue: 0,0:00:45.38,0:00:45.42,Default,,0,0,0,,Vous avez tout à fait raison, M. Xenakis.
-Dialogue: 0,0:00:45.42,0:00:45.62,Default,,0,0,0,,Vous avez tout à fait raison, {\1c&HFF00&\u1}M.{\r} Xenakis.
-Dialogue: 0,0:00:45.62,0:00:45.68,Default,,0,0,0,,Vous avez tout à fait raison, M. Xenakis.
-Dialogue: 0,0:00:45.68,0:00:45.98,Default,,0,0,0,,Vous avez tout à fait raison, M. {\1c&HFF00&\u1}Xenakis.{\r}
-Dialogue: 0,0:00:46.75,0:00:47.13,Default,,0,0,0,,{\1c&HFF00&\u1}Malek{\r} est à l'interne brillant, qui apprend le métier avec moi.
-Dialogue: 0,0:00:47.13,0:00:47.15,Default,,0,0,0,,Malek est à l'interne brillant, qui apprend le métier avec moi.
-Dialogue: 0,0:00:47.15,0:00:47.27,Default,,0,0,0,,Malek {\1c&HFF00&\u1}est{\r} à l'interne brillant, qui apprend le métier avec moi.
-Dialogue: 0,0:00:47.27,0:00:47.31,Default,,0,0,0,,Malek est à l'interne brillant, qui apprend le métier avec moi.
-Dialogue: 0,0:00:47.31,0:00:47.37,Default,,0,0,0,,Malek est {\1c&HFF00&\u1}à{\r} l'interne brillant, qui apprend le métier avec moi.
-Dialogue: 0,0:00:47.37,0:00:47.39,Default,,0,0,0,,Malek est à l'interne brillant, qui apprend le métier avec moi.
-Dialogue: 0,0:00:47.39,0:00:47.75,Default,,0,0,0,,Malek est à {\1c&HFF00&\u1}l'interne{\r} brillant, qui apprend le métier avec moi.
-Dialogue: 0,0:00:47.75,0:00:47.79,Default,,0,0,0,,Malek est à l'interne brillant, qui apprend le métier avec moi.
-Dialogue: 0,0:00:47.79,0:00:48.07,Default,,0,0,0,,Malek est à l'interne {\1c&HFF00&\u1}brillant,{\r} qui apprend le métier avec moi.
-Dialogue: 0,0:00:48.07,0:00:48.11,Default,,0,0,0,,Malek est à l'interne brillant, qui apprend le métier avec moi.
-Dialogue: 0,0:00:48.11,0:00:48.19,Default,,0,0,0,,Malek est à l'interne brillant, {\1c&HFF00&\u1}qui{\r} apprend le métier avec moi.
-Dialogue: 0,0:00:48.19,0:00:48.23,Default,,0,0,0,,Malek est à l'interne brillant, qui apprend le métier avec moi.
-Dialogue: 0,0:00:48.23,0:00:48.44,Default,,0,0,0,,Malek est à l'interne brillant, qui {\1c&HFF00&\u1}apprend{\r} le métier avec moi.
-Dialogue: 0,0:00:48.44,0:00:48.46,Default,,0,0,0,,Malek est à l'interne brillant, qui apprend le métier avec moi.
-Dialogue: 0,0:00:48.46,0:00:48.52,Default,,0,0,0,,Malek est à l'interne brillant, qui apprend {\1c&HFF00&\u1}le{\r} métier avec moi.
-Dialogue: 0,0:00:48.52,0:00:48.54,Default,,0,0,0,,Malek est à l'interne brillant, qui apprend le métier avec moi.
-Dialogue: 0,0:00:48.54,0:00:48.74,Default,,0,0,0,,Malek est à l'interne brillant, qui apprend le {\1c&HFF00&\u1}métier{\r} avec moi.
-Dialogue: 0,0:00:48.74,0:00:48.76,Default,,0,0,0,,Malek est à l'interne brillant, qui apprend le métier avec moi.
-Dialogue: 0,0:00:48.76,0:00:48.88,Default,,0,0,0,,Malek est à l'interne brillant, qui apprend le métier {\1c&HFF00&\u1}avec{\r} moi.
-Dialogue: 0,0:00:48.88,0:00:48.90,Default,,0,0,0,,Malek est à l'interne brillant, qui apprend le métier avec moi.
-Dialogue: 0,0:00:48.90,0:00:49.00,Default,,0,0,0,,Malek est à l'interne brillant, qui apprend le métier avec {\1c&HFF00&\u1}moi.{\r}
-Dialogue: 0,0:00:49.02,0:00:49.06,Default,,0,0,0,,{\1c&HFF00&\u1}Ah!{\r}
-Dialogue: 0,0:00:49.06,0:00:49.20,Default,,0,0,0,,{\1c&HFF00&\u1}C'est{\r} vrai.
-Dialogue: 0,0:00:49.20,0:00:49.22,Default,,0,0,0,,C'est vrai.
-Dialogue: 0,0:00:49.22,0:00:49.30,Default,,0,0,0,,C'est {\1c&HFF00&\u1}vrai.{\r}
-Dialogue: 0,0:00:49.30,0:00:49.44,Default,,0,0,0,,{\1c&HFF00&\u1}Bien.{\r}
-Dialogue: 0,0:00:52.93,0:00:53.21,Default,,0,0,0,,{\1c&HFF00&\u1}Justement,{\r} y a la famille Boboun qui m'attend pour une consultation.
-Dialogue: 0,0:00:53.21,0:00:53.33,Default,,0,0,0,,Justement, y a la famille Boboun qui m'attend pour une consultation.
-Dialogue: 0,0:00:53.33,0:00:53.35,Default,,0,0,0,,Justement, {\1c&HFF00&\u1}y{\r} a la famille Boboun qui m'attend pour une consultation.
-Dialogue: 0,0:00:53.35,0:00:53.37,Default,,0,0,0,,Justement, y a la famille Boboun qui m'attend pour une consultation.
-Dialogue: 0,0:00:53.37,0:00:53.39,Default,,0,0,0,,Justement, y {\1c&HFF00&\u1}a{\r} la famille Boboun qui m'attend pour une consultation.
-Dialogue: 0,0:00:53.39,0:00:53.43,Default,,0,0,0,,Justement, y a la famille Boboun qui m'attend pour une consultation.
-Dialogue: 0,0:00:53.43,0:00:53.49,Default,,0,0,0,,Justement, y a {\1c&HFF00&\u1}la{\r} famille Boboun qui m'attend pour une consultation.
-Dialogue: 0,0:00:53.49,0:00:53.51,Default,,0,0,0,,Justement, y a la famille Boboun qui m'attend pour une consultation.
-Dialogue: 0,0:00:53.51,0:00:53.85,Default,,0,0,0,,Justement, y a la {\1c&HFF00&\u1}famille{\r} Boboun qui m'attend pour une consultation.
-Dialogue: 0,0:00:53.85,0:00:53.89,Default,,0,0,0,,Justement, y a la famille Boboun qui m'attend pour une consultation.
-Dialogue: 0,0:00:53.89,0:00:54.16,Default,,0,0,0,,Justement, y a la famille {\1c&HFF00&\u1}Boboun{\r} qui m'attend pour une consultation.
-Dialogue: 0,0:00:54.16,0:00:54.18,Default,,0,0,0,,Justement, y a la famille Boboun qui m'attend pour une consultation.
-Dialogue: 0,0:00:54.18,0:00:54.26,Default,,0,0,0,,Justement, y a la famille Boboun {\1c&HFF00&\u1}qui{\r} m'attend pour une consultation.
-Dialogue: 0,0:00:54.26,0:00:54.30,Default,,0,0,0,,Justement, y a la famille Boboun qui m'attend pour une consultation.
-Dialogue: 0,0:00:54.30,0:00:54.52,Default,,0,0,0,,Justement, y a la famille Boboun qui {\1c&HFF00&\u1}m'attend{\r} pour une consultation.
-Dialogue: 0,0:00:54.52,0:00:54.54,Default,,0,0,0,,Justement, y a la famille Boboun qui m'attend pour une consultation.
-Dialogue: 0,0:00:54.54,0:00:54.64,Default,,0,0,0,,Justement, y a la famille Boboun qui m'attend {\1c&HFF00&\u1}pour{\r} une consultation.
-Dialogue: 0,0:00:54.64,0:00:54.68,Default,,0,0,0,,Justement, y a la famille Boboun qui m'attend pour une consultation.
-Dialogue: 0,0:00:54.68,0:00:54.90,Default,,0,0,0,,Justement, y a la famille Boboun qui m'attend pour {\1c&HFF00&\u1}une{\r} consultation.
-Dialogue: 0,0:00:54.90,0:00:54.94,Default,,0,0,0,,Justement, y a la famille Boboun qui m'attend pour une consultation.
-Dialogue: 0,0:00:54.94,0:00:55.34,Default,,0,0,0,,Justement, y a la famille Boboun qui m'attend pour une {\1c&HFF00&\u1}consultation.{\r}
-Dialogue: 0,0:00:55.58,0:00:55.64,Default,,0,0,0,,{\1c&HFF00&\u1}Qui?{\r}
-Dialogue: 0,0:00:56.53,0:00:56.79,Default,,0,0,0,,{\1c&HFF00&\u1}Faisons{\r} pas attendre les bobounes, allez.
-Dialogue: 0,0:00:56.79,0:00:56.81,Default,,0,0,0,,Faisons pas attendre les bobounes, allez.
-Dialogue: 0,0:00:56.81,0:00:56.91,Default,,0,0,0,,Faisons {\1c&HFF00&\u1}pas{\r} attendre les bobounes, allez.
-Dialogue: 0,0:00:56.91,0:00:56.93,Default,,0,0,0,,Faisons pas attendre les bobounes, allez.
-Dialogue: 0,0:00:56.93,0:00:57.15,Default,,0,0,0,,Faisons pas {\1c&HFF00&\u1}attendre{\r} les bobounes, allez.
-Dialogue: 0,0:00:57.15,0:00:57.19,Default,,0,0,0,,Faisons pas attendre les bobounes, allez.
-Dialogue: 0,0:00:57.19,0:00:57.25,Default,,0,0,0,,Faisons pas attendre {\1c&HFF00&\u1}les{\r} bobounes, allez.
-Dialogue: 0,0:00:57.25,0:00:57.27,Default,,0,0,0,,Faisons pas attendre les bobounes, allez.
-Dialogue: 0,0:00:57.27,0:00:57.59,Default,,0,0,0,,Faisons pas attendre les {\1c&HFF00&\u1}bobounes,{\r} allez.
-Dialogue: 0,0:00:57.59,0:00:57.61,Default,,0,0,0,,Faisons pas attendre les bobounes, allez.
-Dialogue: 0,0:00:57.61,0:00:57.75,Default,,0,0,0,,Faisons pas attendre les bobounes, {\1c&HFF00&\u1}allez.{\r}

examples/whisperx/sample_fr_01.wav.srt

@@ -1,120 +0,0 @@
-1
-00:00:00,765 --> 00:00:01,309
-Lâchez, c'est bon.
-
-2
-00:00:01,309 --> 00:00:01,891
-Ça va?
-
-3
-00:00:01,911 --> 00:00:01,971
-Oui.
-
-4
-00:00:01,971 --> 00:00:03,495
-Merci beaucoup.
-
-5
-00:00:04,356 --> 00:00:05,037
-Chèque ou espèce?
-
-6
-00:00:06,544 --> 00:00:08,393
-J'ai laissé un chèque sur la commode, il est signé.
-
-7
-00:00:08,393 --> 00:00:09,255
-Vous le remplirez.
-
-8
-00:00:09,275 --> 00:00:09,335
-OK.
-
-9
-00:00:09,355 --> 00:00:09,395
-Oh!
-
-10
-00:00:12,410 --> 00:00:12,590
-Ouh là!
-
-11
-00:00:12,590 --> 00:00:12,731
-Venez.
-
-12
-00:00:14,454 --> 00:00:14,635
-Merci.
-
-13
-00:00:14,655 --> 00:00:14,755
-Ah!
-
-14
-00:00:15,640 --> 00:00:16,626
-C'est qui?
-
-15
-00:00:20,000 --> 00:00:24,234
-C'est pas vrai, qu'est-ce qu'il fout ici, ce con?
-
-16
-00:00:24,515 --> 00:00:26,177
-Excusez-moi, mais je crois que j'ai oublié mon sac chez vous.
-
-17
-00:00:26,177 --> 00:00:26,359
-Ça va?
-
-18
-00:00:31,040 --> 00:00:31,241
-Attendez.
-
-19
-00:00:36,813 --> 00:00:38,278
-Tout à l'heure, là, c'était...
-
-20
-00:00:38,278 --> 00:00:38,359
-Vous?
-
-21
-00:00:39,123 --> 00:00:39,425
-Vous? Pas...
-
-22
-00:00:39,425 --> 00:00:39,706
-Pas lui? Vous?
-
-23
-00:00:44,194 --> 00:00:45,980
-Vous avez tout à fait raison, M. Xenakis.
-
-24
-00:00:46,745 --> 00:00:49,000
-Malek est à l'interne brillant, qui apprend le métier avec moi.
-
-25
-00:00:49,020 --> 00:00:49,061
-Ah!
-
-26
-00:00:49,061 --> 00:00:49,303
-C'est vrai.
-
-27
-00:00:49,303 --> 00:00:49,443
-Bien.
-
-28
-00:00:52,932 --> 00:00:55,338
-Justement, y a la famille Boboun qui m'attend pour une consultation.
-
-29
-00:00:55,581 --> 00:00:55,642
-Qui?
-
-30
-00:00:56,527 --> 00:00:57,753
-Faisons pas attendre les bobounes, allez.
-

examples/whisperx/sample_it_01.wav.ass

@@ -1,177 +0,0 @@
-[Script Info]
-ScriptType: v4.00+
-PlayResX: 384
-PlayResY: 288
-ScaledBorderAndShadow: yes
-
-[V4+ Styles]
-Format: Name, Fontname, Fontsize, PrimaryColour, SecondaryColour, OutlineColour, BackColour, Bold, Italic, Underline, StrikeOut, ScaleX, ScaleY, Spacing, Angle, BorderStyle, Outline, Shadow, Alignment, MarginL, MarginR, MarginV, Encoding
-Style: Default,Arial,24,&Hffffff,&Hffffff,&H0,&H0,0,0,0,0,100,100,0,0,1,1,0,2,10,10,10,0
-
-[Events]
-Format: Layer, Start, End, Style, Name, MarginL, MarginR, MarginV, Effect, Text
-
-Dialogue: 0,0:00:1.20,0:00:1.62,Default,,0,0,0,,{\1c&HFF00&\u1}Signore,{\r} è un piacere, gli amici della vedetta ammirata da tutti noi, questa gemma propria della nostra cultura, saranno naturalmente accolti sotto la mia protezione per la durata del loro soggiorno.
-Dialogue: 0,0:00:1.62,0:00:1.64,Default,,0,0,0,,Signore, è un piacere, gli amici della vedetta ammirata da tutti noi, questa gemma propria della nostra cultura, saranno naturalmente accolti sotto la mia protezione per la durata del loro soggiorno.
-Dialogue: 0,0:00:1.64,0:00:1.66,Default,,0,0,0,,Signore, {\1c&HFF00&\u1}è{\r} un piacere, gli amici della vedetta ammirata da tutti noi, questa gemma propria della nostra cultura, saranno naturalmente accolti sotto la mia protezione per la durata del loro soggiorno.
-Dialogue: 0,0:00:1.66,0:00:1.72,Default,,0,0,0,,Signore, è un piacere, gli amici della vedetta ammirata da tutti noi, questa gemma propria della nostra cultura, saranno naturalmente accolti sotto la mia protezione per la durata del loro soggiorno.
-Dialogue: 0,0:00:1.72,0:00:2.12,Default,,0,0,0,,Signore, è {\1c&HFF00&\u1}un{\r} piacere, gli amici della vedetta ammirata da tutti noi, questa gemma propria della nostra cultura, saranno naturalmente accolti sotto la mia protezione per la durata del loro soggiorno.
-Dialogue: 0,0:00:2.12,0:00:2.18,Default,,0,0,0,,Signore, è un piacere, gli amici della vedetta ammirata da tutti noi, questa gemma propria della nostra cultura, saranno naturalmente accolti sotto la mia protezione per la durata del loro soggiorno.
-Dialogue: 0,0:00:2.18,0:00:2.72,Default,,0,0,0,,Signore, è un {\1c&HFF00&\u1}piacere,{\r} gli amici della vedetta ammirata da tutti noi, questa gemma propria della nostra cultura, saranno naturalmente accolti sotto la mia protezione per la durata del loro soggiorno.
-Dialogue: 0,0:00:2.72,0:00:3.33,Default,,0,0,0,,Signore, è un piacere, gli amici della vedetta ammirata da tutti noi, questa gemma propria della nostra cultura, saranno naturalmente accolti sotto la mia protezione per la durata del loro soggiorno.
-Dialogue: 0,0:00:3.33,0:00:3.45,Default,,0,0,0,,Signore, è un piacere, {\1c&HFF00&\u1}gli{\r} amici della vedetta ammirata da tutti noi, questa gemma propria della nostra cultura, saranno naturalmente accolti sotto la mia protezione per la durata del loro soggiorno.
-Dialogue: 0,0:00:3.45,0:00:3.49,Default,,0,0,0,,Signore, è un piacere, gli amici della vedetta ammirata da tutti noi, questa gemma propria della nostra cultura, saranno naturalmente accolti sotto la mia protezione per la durata del loro soggiorno.
-Dialogue: 0,0:00:3.49,0:00:3.79,Default,,0,0,0,,Signore, è un piacere, gli {\1c&HFF00&\u1}amici{\r} della vedetta ammirata da tutti noi, questa gemma propria della nostra cultura, saranno naturalmente accolti sotto la mia protezione per la durata del loro soggiorno.
-Dialogue: 0,0:00:3.79,0:00:3.83,Default,,0,0,0,,Signore, è un piacere, gli amici della vedetta ammirata da tutti noi, questa gemma propria della nostra cultura, saranno naturalmente accolti sotto la mia protezione per la durata del loro soggiorno.
-Dialogue: 0,0:00:3.83,0:00:4.01,Default,,0,0,0,,Signore, è un piacere, gli amici {\1c&HFF00&\u1}della{\r} vedetta ammirata da tutti noi, questa gemma propria della nostra cultura, saranno naturalmente accolti sotto la mia protezione per la durata del loro soggiorno.
-Dialogue: 0,0:00:4.01,0:00:4.05,Default,,0,0,0,,Signore, è un piacere, gli amici della vedetta ammirata da tutti noi, questa gemma propria della nostra cultura, saranno naturalmente accolti sotto la mia protezione per la durata del loro soggiorno.
-Dialogue: 0,0:00:4.05,0:00:4.35,Default,,0,0,0,,Signore, è un piacere, gli amici della {\1c&HFF00&\u1}vedetta{\r} ammirata da tutti noi, questa gemma propria della nostra cultura, saranno naturalmente accolti sotto la mia protezione per la durata del loro soggiorno.
-Dialogue: 0,0:00:4.35,0:00:4.39,Default,,0,0,0,,Signore, è un piacere, gli amici della vedetta ammirata da tutti noi, questa gemma propria della nostra cultura, saranno naturalmente accolti sotto la mia protezione per la durata del loro soggiorno.
-Dialogue: 0,0:00:4.39,0:00:4.79,Default,,0,0,0,,Signore, è un piacere, gli amici della vedetta {\1c&HFF00&\u1}ammirata{\r} da tutti noi, questa gemma propria della nostra cultura, saranno naturalmente accolti sotto la mia protezione per la durata del loro soggiorno.
-Dialogue: 0,0:00:4.79,0:00:4.85,Default,,0,0,0,,Signore, è un piacere, gli amici della vedetta ammirata da tutti noi, questa gemma propria della nostra cultura, saranno naturalmente accolti sotto la mia protezione per la durata del loro soggiorno.
-Dialogue: 0,0:00:4.85,0:00:4.95,Default,,0,0,0,,Signore, è un piacere, gli amici della vedetta ammirata {\1c&HFF00&\u1}da{\r} tutti noi, questa gemma propria della nostra cultura, saranno naturalmente accolti sotto la mia protezione per la durata del loro soggiorno.
-Dialogue: 0,0:00:4.95,0:00:4.97,Default,,0,0,0,,Signore, è un piacere, gli amici della vedetta ammirata da tutti noi, questa gemma propria della nostra cultura, saranno naturalmente accolti sotto la mia protezione per la durata del loro soggiorno.
-Dialogue: 0,0:00:4.97,0:00:5.15,Default,,0,0,0,,Signore, è un piacere, gli amici della vedetta ammirata da {\1c&HFF00&\u1}tutti{\r} noi, questa gemma propria della nostra cultura, saranno naturalmente accolti sotto la mia protezione per la durata del loro soggiorno.
-Dialogue: 0,0:00:5.15,0:00:5.21,Default,,0,0,0,,Signore, è un piacere, gli amici della vedetta ammirata da tutti noi, questa gemma propria della nostra cultura, saranno naturalmente accolti sotto la mia protezione per la durata del loro soggiorno.
-Dialogue: 0,0:00:5.21,0:00:5.33,Default,,0,0,0,,Signore, è un piacere, gli amici della vedetta ammirata da tutti {\1c&HFF00&\u1}noi,{\r} questa gemma propria della nostra cultura, saranno naturalmente accolti sotto la mia protezione per la durata del loro soggiorno.
-Dialogue: 0,0:00:5.33,0:00:5.41,Default,,0,0,0,,Signore, è un piacere, gli amici della vedetta ammirata da tutti noi, questa gemma propria della nostra cultura, saranno naturalmente accolti sotto la mia protezione per la durata del loro soggiorno.
-Dialogue: 0,0:00:5.41,0:00:5.61,Default,,0,0,0,,Signore, è un piacere, gli amici della vedetta ammirata da tutti noi, {\1c&HFF00&\u1}questa{\r} gemma propria della nostra cultura, saranno naturalmente accolti sotto la mia protezione per la durata del loro soggiorno.
-Dialogue: 0,0:00:5.61,0:00:5.79,Default,,0,0,0,,Signore, è un piacere, gli amici della vedetta ammirata da tutti noi, questa gemma propria della nostra cultura, saranno naturalmente accolti sotto la mia protezione per la durata del loro soggiorno.
-Dialogue: 0,0:00:5.79,0:00:6.07,Default,,0,0,0,,Signore, è un piacere, gli amici della vedetta ammirata da tutti noi, questa {\1c&HFF00&\u1}gemma{\r} propria della nostra cultura, saranno naturalmente accolti sotto la mia protezione per la durata del loro soggiorno.
-Dialogue: 0,0:00:6.07,0:00:6.13,Default,,0,0,0,,Signore, è un piacere, gli amici della vedetta ammirata da tutti noi, questa gemma propria della nostra cultura, saranno naturalmente accolti sotto la mia protezione per la durata del loro soggiorno.
-Dialogue: 0,0:00:6.13,0:00:6.51,Default,,0,0,0,,Signore, è un piacere, gli amici della vedetta ammirata da tutti noi, questa gemma {\1c&HFF00&\u1}propria{\r} della nostra cultura, saranno naturalmente accolti sotto la mia protezione per la durata del loro soggiorno.
-Dialogue: 0,0:00:6.51,0:00:6.57,Default,,0,0,0,,Signore, è un piacere, gli amici della vedetta ammirata da tutti noi, questa gemma propria della nostra cultura, saranno naturalmente accolti sotto la mia protezione per la durata del loro soggiorno.
-Dialogue: 0,0:00:6.57,0:00:6.77,Default,,0,0,0,,Signore, è un piacere, gli amici della vedetta ammirata da tutti noi, questa gemma propria {\1c&HFF00&\u1}della{\r} nostra cultura, saranno naturalmente accolti sotto la mia protezione per la durata del loro soggiorno.
-Dialogue: 0,0:00:6.77,0:00:6.81,Default,,0,0,0,,Signore, è un piacere, gli amici della vedetta ammirata da tutti noi, questa gemma propria della nostra cultura, saranno naturalmente accolti sotto la mia protezione per la durata del loro soggiorno.
-Dialogue: 0,0:00:6.81,0:00:6.99,Default,,0,0,0,,Signore, è un piacere, gli amici della vedetta ammirata da tutti noi, questa gemma propria della {\1c&HFF00&\u1}nostra{\r} cultura, saranno naturalmente accolti sotto la mia protezione per la durata del loro soggiorno.
-Dialogue: 0,0:00:6.99,0:00:7.07,Default,,0,0,0,,Signore, è un piacere, gli amici della vedetta ammirata da tutti noi, questa gemma propria della nostra cultura, saranno naturalmente accolti sotto la mia protezione per la durata del loro soggiorno.
-Dialogue: 0,0:00:7.07,0:00:7.35,Default,,0,0,0,,Signore, è un piacere, gli amici della vedetta ammirata da tutti noi, questa gemma propria della nostra {\1c&HFF00&\u1}cultura,{\r} saranno naturalmente accolti sotto la mia protezione per la durata del loro soggiorno.
-Dialogue: 0,0:00:7.35,0:00:7.41,Default,,0,0,0,,Signore, è un piacere, gli amici della vedetta ammirata da tutti noi, questa gemma propria della nostra cultura, saranno naturalmente accolti sotto la mia protezione per la durata del loro soggiorno.
-Dialogue: 0,0:00:7.41,0:00:7.73,Default,,0,0,0,,Signore, è un piacere, gli amici della vedetta ammirata da tutti noi, questa gemma propria della nostra cultura, {\1c&HFF00&\u1}saranno{\r} naturalmente accolti sotto la mia protezione per la durata del loro soggiorno.
-Dialogue: 0,0:00:7.73,0:00:7.87,Default,,0,0,0,,Signore, è un piacere, gli amici della vedetta ammirata da tutti noi, questa gemma propria della nostra cultura, saranno naturalmente accolti sotto la mia protezione per la durata del loro soggiorno.
-Dialogue: 0,0:00:7.87,0:00:8.47,Default,,0,0,0,,Signore, è un piacere, gli amici della vedetta ammirata da tutti noi, questa gemma propria della nostra cultura, saranno {\1c&HFF00&\u1}naturalmente{\r} accolti sotto la mia protezione per la durata del loro soggiorno.
-Dialogue: 0,0:00:8.47,0:00:8.55,Default,,0,0,0,,Signore, è un piacere, gli amici della vedetta ammirata da tutti noi, questa gemma propria della nostra cultura, saranno naturalmente accolti sotto la mia protezione per la durata del loro soggiorno.
-Dialogue: 0,0:00:8.55,0:00:8.85,Default,,0,0,0,,Signore, è un piacere, gli amici della vedetta ammirata da tutti noi, questa gemma propria della nostra cultura, saranno naturalmente {\1c&HFF00&\u1}accolti{\r} sotto la mia protezione per la durata del loro soggiorno.
-Dialogue: 0,0:00:8.85,0:00:8.91,Default,,0,0,0,,Signore, è un piacere, gli amici della vedetta ammirata da tutti noi, questa gemma propria della nostra cultura, saranno naturalmente accolti sotto la mia protezione per la durata del loro soggiorno.
-Dialogue: 0,0:00:8.91,0:00:9.07,Default,,0,0,0,,Signore, è un piacere, gli amici della vedetta ammirata da tutti noi, questa gemma propria della nostra cultura, saranno naturalmente accolti {\1c&HFF00&\u1}sotto{\r} la mia protezione per la durata del loro soggiorno.
-Dialogue: 0,0:00:9.07,0:00:9.13,Default,,0,0,0,,Signore, è un piacere, gli amici della vedetta ammirata da tutti noi, questa gemma propria della nostra cultura, saranno naturalmente accolti sotto la mia protezione per la durata del loro soggiorno.
-Dialogue: 0,0:00:9.13,0:00:9.19,Default,,0,0,0,,Signore, è un piacere, gli amici della vedetta ammirata da tutti noi, questa gemma propria della nostra cultura, saranno naturalmente accolti sotto {\1c&HFF00&\u1}la{\r} mia protezione per la durata del loro soggiorno.
-Dialogue: 0,0:00:9.19,0:00:9.23,Default,,0,0,0,,Signore, è un piacere, gli amici della vedetta ammirata da tutti noi, questa gemma propria della nostra cultura, saranno naturalmente accolti sotto la mia protezione per la durata del loro soggiorno.
-Dialogue: 0,0:00:9.23,0:00:9.33,Default,,0,0,0,,Signore, è un piacere, gli amici della vedetta ammirata da tutti noi, questa gemma propria della nostra cultura, saranno naturalmente accolti sotto la {\1c&HFF00&\u1}mia{\r} protezione per la durata del loro soggiorno.
-Dialogue: 0,0:00:9.33,0:00:9.37,Default,,0,0,0,,Signore, è un piacere, gli amici della vedetta ammirata da tutti noi, questa gemma propria della nostra cultura, saranno naturalmente accolti sotto la mia protezione per la durata del loro soggiorno.
-Dialogue: 0,0:00:9.37,0:00:9.82,Default,,0,0,0,,Signore, è un piacere, gli amici della vedetta ammirata da tutti noi, questa gemma propria della nostra cultura, saranno naturalmente accolti sotto la mia {\1c&HFF00&\u1}protezione{\r} per la durata del loro soggiorno.
-Dialogue: 0,0:00:9.82,0:00:9.88,Default,,0,0,0,,Signore, è un piacere, gli amici della vedetta ammirata da tutti noi, questa gemma propria della nostra cultura, saranno naturalmente accolti sotto la mia protezione per la durata del loro soggiorno.
-Dialogue: 0,0:00:9.88,0:00:9.96,Default,,0,0,0,,Signore, è un piacere, gli amici della vedetta ammirata da tutti noi, questa gemma propria della nostra cultura, saranno naturalmente accolti sotto la mia protezione {\1c&HFF00&\u1}per{\r} la durata del loro soggiorno.
-Dialogue: 0,0:00:9.96,0:00:10.02,Default,,0,0,0,,Signore, è un piacere, gli amici della vedetta ammirata da tutti noi, questa gemma propria della nostra cultura, saranno naturalmente accolti sotto la mia protezione per la durata del loro soggiorno.
-Dialogue: 0,0:00:10.02,0:00:10.08,Default,,0,0,0,,Signore, è un piacere, gli amici della vedetta ammirata da tutti noi, questa gemma propria della nostra cultura, saranno naturalmente accolti sotto la mia protezione per {\1c&HFF00&\u1}la{\r} durata del loro soggiorno.
-Dialogue: 0,0:00:10.08,0:00:10.12,Default,,0,0,0,,Signore, è un piacere, gli amici della vedetta ammirata da tutti noi, questa gemma propria della nostra cultura, saranno naturalmente accolti sotto la mia protezione per la durata del loro soggiorno.
-Dialogue: 0,0:00:10.12,0:00:10.44,Default,,0,0,0,,Signore, è un piacere, gli amici della vedetta ammirata da tutti noi, questa gemma propria della nostra cultura, saranno naturalmente accolti sotto la mia protezione per la {\1c&HFF00&\u1}durata{\r} del loro soggiorno.
-Dialogue: 0,0:00:10.44,0:00:10.50,Default,,0,0,0,,Signore, è un piacere, gli amici della vedetta ammirata da tutti noi, questa gemma propria della nostra cultura, saranno naturalmente accolti sotto la mia protezione per la durata del loro soggiorno.
-Dialogue: 0,0:00:10.50,0:00:10.60,Default,,0,0,0,,Signore, è un piacere, gli amici della vedetta ammirata da tutti noi, questa gemma propria della nostra cultura, saranno naturalmente accolti sotto la mia protezione per la durata {\1c&HFF00&\u1}del{\r} loro soggiorno.
-Dialogue: 0,0:00:10.60,0:00:10.62,Default,,0,0,0,,Signore, è un piacere, gli amici della vedetta ammirata da tutti noi, questa gemma propria della nostra cultura, saranno naturalmente accolti sotto la mia protezione per la durata del loro soggiorno.
-Dialogue: 0,0:00:10.62,0:00:10.78,Default,,0,0,0,,Signore, è un piacere, gli amici della vedetta ammirata da tutti noi, questa gemma propria della nostra cultura, saranno naturalmente accolti sotto la mia protezione per la durata del {\1c&HFF00&\u1}loro{\r} soggiorno.
-Dialogue: 0,0:00:10.78,0:00:10.86,Default,,0,0,0,,Signore, è un piacere, gli amici della vedetta ammirata da tutti noi, questa gemma propria della nostra cultura, saranno naturalmente accolti sotto la mia protezione per la durata del loro soggiorno.
-Dialogue: 0,0:00:10.86,0:00:11.30,Default,,0,0,0,,Signore, è un piacere, gli amici della vedetta ammirata da tutti noi, questa gemma propria della nostra cultura, saranno naturalmente accolti sotto la mia protezione per la durata del loro {\1c&HFF00&\u1}soggiorno.{\r}
-Dialogue: 0,0:00:13.00,0:00:13.12,Default,,0,0,0,,{\1c&HFF00&\u1}Grazie.{\r}
-Dialogue: 0,0:00:15.60,0:00:17.89,Default,,0,0,0,,{\1c&HFF00&\u1}Gorlami?{\r} Lo pronuncio correttamente?
-Dialogue: 0,0:00:17.89,0:00:18.75,Default,,0,0,0,,Gorlami? Lo pronuncio correttamente?
-Dialogue: 0,0:00:18.75,0:00:18.83,Default,,0,0,0,,Gorlami? {\1c&HFF00&\u1}Lo{\r} pronuncio correttamente?
-Dialogue: 0,0:00:18.83,0:00:18.87,Default,,0,0,0,,Gorlami? Lo pronuncio correttamente?
-Dialogue: 0,0:00:18.87,0:00:19.31,Default,,0,0,0,,Gorlami? Lo {\1c&HFF00&\u1}pronuncio{\r} correttamente?
-Dialogue: 0,0:00:19.31,0:00:19.37,Default,,0,0,0,,Gorlami? Lo pronuncio correttamente?
-Dialogue: 0,0:00:19.37,0:00:19.87,Default,,0,0,0,,Gorlami? Lo pronuncio {\1c&HFF00&\u1}correttamente?{\r}
-Dialogue: 0,0:00:21.58,0:00:21.74,Default,,0,0,0,,{\1c&HFF00&\u1}Sì,{\r} corretto.
-Dialogue: 0,0:00:21.74,0:00:22.34,Default,,0,0,0,,Sì, corretto.
-Dialogue: 0,0:00:22.34,0:00:22.72,Default,,0,0,0,,Sì, {\1c&HFF00&\u1}corretto.{\r}
-Dialogue: 0,0:00:23.54,0:00:24.82,Default,,0,0,0,,{\1c&HFF00&\u1}Gorlami?{\r} Per cortesia, me lo ripeto ancora.
-Dialogue: 0,0:00:24.82,0:00:25.45,Default,,0,0,0,,Gorlami? Per cortesia, me lo ripeto ancora.
-Dialogue: 0,0:00:25.45,0:00:25.53,Default,,0,0,0,,Gorlami? {\1c&HFF00&\u1}Per{\r} cortesia, me lo ripeto ancora.
-Dialogue: 0,0:00:25.53,0:00:25.61,Default,,0,0,0,,Gorlami? Per cortesia, me lo ripeto ancora.
-Dialogue: 0,0:00:25.61,0:00:26.17,Default,,0,0,0,,Gorlami? Per {\1c&HFF00&\u1}cortesia,{\r} me lo ripeto ancora.
-Dialogue: 0,0:00:26.17,0:00:26.39,Default,,0,0,0,,Gorlami? Per cortesia, me lo ripeto ancora.
-Dialogue: 0,0:00:26.39,0:00:26.45,Default,,0,0,0,,Gorlami? Per cortesia, {\1c&HFF00&\u1}me{\r} lo ripeto ancora.
-Dialogue: 0,0:00:26.45,0:00:26.49,Default,,0,0,0,,Gorlami? Per cortesia, me lo ripeto ancora.
-Dialogue: 0,0:00:26.49,0:00:26.55,Default,,0,0,0,,Gorlami? Per cortesia, me {\1c&HFF00&\u1}lo{\r} ripeto ancora.
-Dialogue: 0,0:00:26.55,0:00:26.61,Default,,0,0,0,,Gorlami? Per cortesia, me lo ripeto ancora.
-Dialogue: 0,0:00:26.61,0:00:27.05,Default,,0,0,0,,Gorlami? Per cortesia, me lo {\1c&HFF00&\u1}ripeto{\r} ancora.
-Dialogue: 0,0:00:27.05,0:00:27.11,Default,,0,0,0,,Gorlami? Per cortesia, me lo ripeto ancora.
-Dialogue: 0,0:00:27.11,0:00:27.49,Default,,0,0,0,,Gorlami? Per cortesia, me lo ripeto {\1c&HFF00&\u1}ancora.{\r}
-Dialogue: 0,0:00:27.55,0:00:28.79,Default,,0,0,0,,{\1c&HFF00&\u1}ancora{\r} gourlami scusi con me gourlami ancora una volta gourlami e come si chiama lei antonio
-Dialogue: 0,0:00:28.79,0:00:28.83,Default,,0,0,0,,ancora gourlami scusi con me gourlami ancora una volta gourlami e come si chiama lei antonio
-Dialogue: 0,0:00:28.83,0:00:30.96,Default,,0,0,0,,ancora {\1c&HFF00&\u1}gourlami{\r} scusi con me gourlami ancora una volta gourlami e come si chiama lei antonio
-Dialogue: 0,0:00:30.96,0:00:31.02,Default,,0,0,0,,ancora gourlami scusi con me gourlami ancora una volta gourlami e come si chiama lei antonio
-Dialogue: 0,0:00:31.02,0:00:31.36,Default,,0,0,0,,ancora gourlami {\1c&HFF00&\u1}scusi{\r} con me gourlami ancora una volta gourlami e come si chiama lei antonio
-Dialogue: 0,0:00:31.36,0:00:31.46,Default,,0,0,0,,ancora gourlami scusi con me gourlami ancora una volta gourlami e come si chiama lei antonio
-Dialogue: 0,0:00:31.46,0:00:31.66,Default,,0,0,0,,ancora gourlami scusi {\1c&HFF00&\u1}con{\r} me gourlami ancora una volta gourlami e come si chiama lei antonio
-Dialogue: 0,0:00:31.66,0:00:31.72,Default,,0,0,0,,ancora gourlami scusi con me gourlami ancora una volta gourlami e come si chiama lei antonio
-Dialogue: 0,0:00:31.72,0:00:31.88,Default,,0,0,0,,ancora gourlami scusi con {\1c&HFF00&\u1}me{\r} gourlami ancora una volta gourlami e come si chiama lei antonio
-Dialogue: 0,0:00:31.88,0:00:33.46,Default,,0,0,0,,ancora gourlami scusi con me gourlami ancora una volta gourlami e come si chiama lei antonio
-Dialogue: 0,0:00:33.46,0:00:34.12,Default,,0,0,0,,ancora gourlami scusi con me {\1c&HFF00&\u1}gourlami{\r} ancora una volta gourlami e come si chiama lei antonio
-Dialogue: 0,0:00:34.12,0:00:34.84,Default,,0,0,0,,ancora gourlami scusi con me gourlami ancora una volta gourlami e come si chiama lei antonio
-Dialogue: 0,0:00:34.84,0:00:35.20,Default,,0,0,0,,ancora gourlami scusi con me gourlami {\1c&HFF00&\u1}ancora{\r} una volta gourlami e come si chiama lei antonio
-Dialogue: 0,0:00:35.20,0:00:35.32,Default,,0,0,0,,ancora gourlami scusi con me gourlami ancora una volta gourlami e come si chiama lei antonio
-Dialogue: 0,0:00:35.32,0:00:35.44,Default,,0,0,0,,ancora gourlami scusi con me gourlami ancora {\1c&HFF00&\u1}una{\r} volta gourlami e come si chiama lei antonio
-Dialogue: 0,0:00:35.44,0:00:35.48,Default,,0,0,0,,ancora gourlami scusi con me gourlami ancora una volta gourlami e come si chiama lei antonio
-Dialogue: 0,0:00:35.48,0:00:35.82,Default,,0,0,0,,ancora gourlami scusi con me gourlami ancora una {\1c&HFF00&\u1}volta{\r} gourlami e come si chiama lei antonio
-Dialogue: 0,0:00:35.82,0:00:39.16,Default,,0,0,0,,ancora gourlami scusi con me gourlami ancora una volta gourlami e come si chiama lei antonio
-Dialogue: 0,0:00:39.16,0:00:39.72,Default,,0,0,0,,ancora gourlami scusi con me gourlami ancora una volta {\1c&HFF00&\u1}gourlami{\r} e come si chiama lei antonio
-Dialogue: 0,0:00:39.72,0:00:40.96,Default,,0,0,0,,ancora gourlami scusi con me gourlami ancora una volta gourlami e come si chiama lei antonio
-Dialogue: 0,0:00:40.96,0:00:41.14,Default,,0,0,0,,ancora gourlami scusi con me gourlami ancora una volta gourlami {\1c&HFF00&\u1}e{\r} come si chiama lei antonio
-Dialogue: 0,0:00:41.14,0:00:41.20,Default,,0,0,0,,ancora gourlami scusi con me gourlami ancora una volta gourlami e come si chiama lei antonio
-Dialogue: 0,0:00:41.20,0:00:41.34,Default,,0,0,0,,ancora gourlami scusi con me gourlami ancora una volta gourlami e {\1c&HFF00&\u1}come{\r} si chiama lei antonio
-Dialogue: 0,0:00:41.34,0:00:41.38,Default,,0,0,0,,ancora gourlami scusi con me gourlami ancora una volta gourlami e come si chiama lei antonio
-Dialogue: 0,0:00:41.38,0:00:41.46,Default,,0,0,0,,ancora gourlami scusi con me gourlami ancora una volta gourlami e come {\1c&HFF00&\u1}si{\r} chiama lei antonio
-Dialogue: 0,0:00:41.46,0:00:41.50,Default,,0,0,0,,ancora gourlami scusi con me gourlami ancora una volta gourlami e come si chiama lei antonio
-Dialogue: 0,0:00:41.50,0:00:41.70,Default,,0,0,0,,ancora gourlami scusi con me gourlami ancora una volta gourlami e come si {\1c&HFF00&\u1}chiama{\r} lei antonio
-Dialogue: 0,0:00:41.70,0:00:41.80,Default,,0,0,0,,ancora gourlami scusi con me gourlami ancora una volta gourlami e come si chiama lei antonio
-Dialogue: 0,0:00:41.80,0:00:42.06,Default,,0,0,0,,ancora gourlami scusi con me gourlami ancora una volta gourlami e come si chiama {\1c&HFF00&\u1}lei{\r} antonio
-Dialogue: 0,0:00:42.06,0:00:43.44,Default,,0,0,0,,ancora gourlami scusi con me gourlami ancora una volta gourlami e come si chiama lei antonio
-Dialogue: 0,0:00:43.44,0:00:43.98,Default,,0,0,0,,ancora gourlami scusi con me gourlami ancora una volta gourlami e come si chiama lei {\1c&HFF00&\u1}antonio{\r}
-Dialogue: 0,0:00:44.36,0:00:45.14,Default,,0,0,0,,{\1c&HFF00&\u1}margarete{\r} ancora margarete un'altra volta ma adesso vorrei proprio sentire la musica
-Dialogue: 0,0:00:45.14,0:00:46.56,Default,,0,0,0,,margarete ancora margarete un'altra volta ma adesso vorrei proprio sentire la musica
-Dialogue: 0,0:00:46.56,0:00:46.85,Default,,0,0,0,,margarete {\1c&HFF00&\u1}ancora{\r} margarete un'altra volta ma adesso vorrei proprio sentire la musica
-Dialogue: 0,0:00:46.85,0:00:47.85,Default,,0,0,0,,margarete ancora margarete un'altra volta ma adesso vorrei proprio sentire la musica
-Dialogue: 0,0:00:47.85,0:00:49.15,Default,,0,0,0,,margarete ancora {\1c&HFF00&\u1}margarete{\r} un'altra volta ma adesso vorrei proprio sentire la musica
-Dialogue: 0,0:00:49.15,0:00:49.43,Default,,0,0,0,,margarete ancora margarete un'altra volta ma adesso vorrei proprio sentire la musica
-Dialogue: 0,0:00:49.43,0:00:49.85,Default,,0,0,0,,margarete ancora margarete {\1c&HFF00&\u1}un'altra{\r} volta ma adesso vorrei proprio sentire la musica
-Dialogue: 0,0:00:49.85,0:00:49.91,Default,,0,0,0,,margarete ancora margarete un'altra volta ma adesso vorrei proprio sentire la musica
-Dialogue: 0,0:00:49.91,0:00:50.13,Default,,0,0,0,,margarete ancora margarete un'altra {\1c&HFF00&\u1}volta{\r} ma adesso vorrei proprio sentire la musica
-Dialogue: 0,0:00:50.13,0:00:50.19,Default,,0,0,0,,margarete ancora margarete un'altra volta ma adesso vorrei proprio sentire la musica
-Dialogue: 0,0:00:50.19,0:00:50.37,Default,,0,0,0,,margarete ancora margarete un'altra volta {\1c&HFF00&\u1}ma{\r} adesso vorrei proprio sentire la musica
-Dialogue: 0,0:00:50.37,0:00:50.43,Default,,0,0,0,,margarete ancora margarete un'altra volta ma adesso vorrei proprio sentire la musica
-Dialogue: 0,0:00:50.43,0:00:50.73,Default,,0,0,0,,margarete ancora margarete un'altra volta ma {\1c&HFF00&\u1}adesso{\r} vorrei proprio sentire la musica
-Dialogue: 0,0:00:50.73,0:00:50.77,Default,,0,0,0,,margarete ancora margarete un'altra volta ma adesso vorrei proprio sentire la musica
-Dialogue: 0,0:00:50.77,0:00:51.01,Default,,0,0,0,,margarete ancora margarete un'altra volta ma adesso {\1c&HFF00&\u1}vorrei{\r} proprio sentire la musica
-Dialogue: 0,0:00:51.01,0:00:51.05,Default,,0,0,0,,margarete ancora margarete un'altra volta ma adesso vorrei proprio sentire la musica
-Dialogue: 0,0:00:51.05,0:00:51.29,Default,,0,0,0,,margarete ancora margarete un'altra volta ma adesso vorrei {\1c&HFF00&\u1}proprio{\r} sentire la musica
-Dialogue: 0,0:00:51.29,0:00:51.43,Default,,0,0,0,,margarete ancora margarete un'altra volta ma adesso vorrei proprio sentire la musica
-Dialogue: 0,0:00:51.43,0:00:51.71,Default,,0,0,0,,margarete ancora margarete un'altra volta ma adesso vorrei proprio {\1c&HFF00&\u1}sentire{\r} la musica
-Dialogue: 0,0:00:51.71,0:00:51.79,Default,,0,0,0,,margarete ancora margarete un'altra volta ma adesso vorrei proprio sentire la musica
-Dialogue: 0,0:00:51.79,0:00:51.88,Default,,0,0,0,,margarete ancora margarete un'altra volta ma adesso vorrei proprio sentire {\1c&HFF00&\u1}la{\r} musica
-Dialogue: 0,0:00:51.88,0:00:51.92,Default,,0,0,0,,margarete ancora margarete un'altra volta ma adesso vorrei proprio sentire la musica
-Dialogue: 0,0:00:51.92,0:00:52.36,Default,,0,0,0,,margarete ancora margarete un'altra volta ma adesso vorrei proprio sentire la {\1c&HFF00&\u1}musica{\r}
-Dialogue: 0,0:00:52.36,0:00:52.40,Default,,0,0,0,,{\1c&HFF00&\u1}le{\r} parole margheriti margheriti e lei dominic decoco come dominic decoco bravo bravo
-Dialogue: 0,0:00:52.40,0:00:52.52,Default,,0,0,0,,le parole margheriti margheriti e lei dominic decoco come dominic decoco bravo bravo
-Dialogue: 0,0:00:52.52,0:00:54.32,Default,,0,0,0,,le {\1c&HFF00&\u1}parole{\r} margheriti margheriti e lei dominic decoco come dominic decoco bravo bravo
-Dialogue: 0,0:00:54.32,0:00:54.40,Default,,0,0,0,,le parole margheriti margheriti e lei dominic decoco come dominic decoco bravo bravo
-Dialogue: 0,0:00:54.40,0:00:56.40,Default,,0,0,0,,le parole {\1c&HFF00&\u1}margheriti{\r} margheriti e lei dominic decoco come dominic decoco bravo bravo
-Dialogue: 0,0:00:56.40,0:00:56.96,Default,,0,0,0,,le parole margheriti margheriti e lei dominic decoco come dominic decoco bravo bravo
-Dialogue: 0,0:00:56.96,0:00:57.92,Default,,0,0,0,,le parole margheriti {\1c&HFF00&\u1}margheriti{\r} e lei dominic decoco come dominic decoco bravo bravo
-Dialogue: 0,0:00:57.92,0:00:58.53,Default,,0,0,0,,le parole margheriti margheriti e lei dominic decoco come dominic decoco bravo bravo
-Dialogue: 0,0:00:58.53,0:00:58.59,Default,,0,0,0,,le parole margheriti margheriti {\1c&HFF00&\u1}e{\r} lei dominic decoco come dominic decoco bravo bravo
-Dialogue: 0,0:00:58.59,0:00:58.65,Default,,0,0,0,,le parole margheriti margheriti e lei dominic decoco come dominic decoco bravo bravo
-Dialogue: 0,0:00:58.65,0:00:58.97,Default,,0,0,0,,le parole margheriti margheriti e {\1c&HFF00&\u1}lei{\r} dominic decoco come dominic decoco bravo bravo
-Dialogue: 0,0:00:58.97,0:00:59.59,Default,,0,0,0,,le parole margheriti margheriti e lei dominic decoco come dominic decoco bravo bravo
-Dialogue: 0,0:00:59.59,0:00:59.91,Default,,0,0,0,,le parole margheriti margheriti e lei {\1c&HFF00&\u1}dominic{\r} decoco come dominic decoco bravo bravo
-Dialogue: 0,0:00:59.91,0:00:59.95,Default,,0,0,0,,le parole margheriti margheriti e lei dominic decoco come dominic decoco bravo bravo
-Dialogue: 0,0:00:59.95,0:01:0.35,Default,,0,0,0,,le parole margheriti margheriti e lei dominic {\1c&HFF00&\u1}decoco{\r} come dominic decoco bravo bravo
-Dialogue: 0,0:01:0.35,0:01:0.55,Default,,0,0,0,,le parole margheriti margheriti e lei dominic decoco come dominic decoco bravo bravo
-Dialogue: 0,0:01:0.55,0:01:0.73,Default,,0,0,0,,le parole margheriti margheriti e lei dominic decoco {\1c&HFF00&\u1}come{\r} dominic decoco bravo bravo
-Dialogue: 0,0:01:0.73,0:01:1.25,Default,,0,0,0,,le parole margheriti margheriti e lei dominic decoco come dominic decoco bravo bravo
-Dialogue: 0,0:01:1.25,0:01:1.57,Default,,0,0,0,,le parole margheriti margheriti e lei dominic decoco come {\1c&HFF00&\u1}dominic{\r} decoco bravo bravo
-Dialogue: 0,0:01:1.57,0:01:1.61,Default,,0,0,0,,le parole margheriti margheriti e lei dominic decoco come dominic decoco bravo bravo
-Dialogue: 0,0:01:1.61,0:01:2.01,Default,,0,0,0,,le parole margheriti margheriti e lei dominic decoco come dominic {\1c&HFF00&\u1}decoco{\r} bravo bravo
-Dialogue: 0,0:01:2.01,0:01:2.17,Default,,0,0,0,,le parole margheriti margheriti e lei dominic decoco come dominic decoco bravo bravo
-Dialogue: 0,0:01:2.17,0:01:2.45,Default,,0,0,0,,le parole margheriti margheriti e lei dominic decoco come dominic decoco {\1c&HFF00&\u1}bravo{\r} bravo
-Dialogue: 0,0:01:2.45,0:01:2.91,Default,,0,0,0,,le parole margheriti margheriti e lei dominic decoco come dominic decoco bravo bravo
-Dialogue: 0,0:01:2.91,0:01:3.29,Default,,0,0,0,,le parole margheriti margheriti e lei dominic decoco come dominic decoco bravo {\1c&HFF00&\u1}bravo{\r}

examples/whisperx/sample_it_01.wav.srt

@@ -1,32 +0,0 @@
-1
-00:00:01,202 --> 00:00:11,297
-Signore, è un piacere, gli amici della vedetta ammirata da tutti noi, questa gemma propria della nostra cultura, saranno naturalmente accolti sotto la mia protezione per la durata del loro soggiorno.
-
-2
-00:00:13,000 --> 00:00:13,120
-Grazie.
-
-3
-00:00:15,602 --> 00:00:19,874
-Gorlami? Lo pronuncio correttamente?
-
-4
-00:00:21,576 --> 00:00:22,717
-Sì, corretto.
-
-5
-00:00:23,540 --> 00:00:27,495
-Gorlami? Per cortesia, me lo ripeto ancora.
-
-6
-00:00:27,555 --> 00:00:43,979
-ancora gourlami scusi con me gourlami ancora una volta gourlami e come si chiama lei antonio
-
-7
-00:00:44,360 --> 00:00:52,356
-margarete ancora margarete un'altra volta ma adesso vorrei proprio sentire la musica
-
-8
-00:00:52,356 --> 00:01:03,292
-le parole margheriti margheriti e lei dominic decoco come dominic decoco bravo bravo
-

pyproject.toml

@@ -0,0 +1,36 @@
+[project]
+urls = { repository = "https://github.com/m-bain/whisperx" }
+authors = [{ name = "Max Bain" }]
+name = "whisperx"
+version = "3.3.4"
+description = "Time-Accurate Automatic Speech Recognition using Whisper."
+readme = "README.md"
+requires-python = ">=3.9, <3.13"
+license = { text = "BSD-2-Clause" }
+
+dependencies = [
+    "ctranslate2<4.5.0",
+    "faster-whisper>=1.1.1",
+    "nltk>=3.9.1",
+    "numpy>=2.0.2",
+    "onnxruntime>=1.19,<1.20.0",
+    "pandas>=2.2.3",
+    "pyannote-audio>=3.3.2",
+    "torch<2.4.0",
+    "torchaudio",
+    "transformers>=4.48.0",
+]
+
+
+[project.scripts]
+whisperx = "whisperx.__main__:cli"
+
+[build-system]
+requires = ["setuptools"]
+
+[tool.setuptools]
+include-package-data = true
+
+[tool.setuptools.packages.find]
+where = ["."]
+include = ["whisperx*"]

requirements.txt

@@ -1,9 +0,0 @@
-numpy
-torch
-torchaudio
-tqdm
-soundfile
-more-itertools
-transformers>=4.19.0
-ffmpeg-python==0.2.0
-pyannote.audio

setup.py

@@ -1,28 +0,0 @@
-import os
-
-import pkg_resources
-from setuptools import setup, find_packages
-
-setup(
-    name="whisperx",
-    py_modules=["whisperx"],
-    version="1.0",
-    description="Time-Accurate Automatic Speech Recognition using Whisper.",
-    readme="README.md",
-    python_requires=">=3.7",
-    author="Max Bain",
-    url="https://github.com/m-bain/whisperx",
-    license="MIT",
-    packages=find_packages(exclude=["tests*"]),
-    install_requires=[
-        str(r)
-        for r in pkg_resources.parse_requirements(
-            open(os.path.join(os.path.dirname(__file__), "requirements.txt"))
-        )
-    ],
-    entry_points = {
-        'console_scripts': ['whisperx=whisperx.transcribe:cli'],
-    },
-    include_package_data=True,
-    extras_require={'dev': ['pytest']},
-)

tests/test_audio.py

@@ -1,19 +0,0 @@
-import os.path
-
-import numpy as np
-
-from whisper.audio import load_audio, log_mel_spectrogram, SAMPLE_RATE
-
-
-def test_audio():
-    audio_path = os.path.join(os.path.dirname(__file__), "jfk.flac")
-    audio = load_audio(audio_path)
-    assert audio.ndim == 1
-    assert SAMPLE_RATE * 10 < audio.shape[0] < SAMPLE_RATE * 12
-    assert 0 < audio.std() < 1
-
-    mel_from_audio = log_mel_spectrogram(audio)
-    mel_from_file = log_mel_spectrogram(audio_path)
-
-    assert np.allclose(mel_from_audio, mel_from_file)
-    assert mel_from_audio.max() - mel_from_audio.min() <= 2.0

tests/test_normalizer.py

@@ -1,92 +0,0 @@
-import pytest
-
-from whisper.normalizers import EnglishTextNormalizer
-from whisper.normalizers.english import EnglishNumberNormalizer, EnglishSpellingNormalizer
-
-
-@pytest.mark.parametrize("std", [EnglishNumberNormalizer(), EnglishTextNormalizer()])
-def test_number_normalizer(std):
-    assert std("two") == "2"
-    assert std("thirty one") == "31"
-    assert std("five twenty four") == "524"
-    assert std("nineteen ninety nine") == "1999"
-    assert std("twenty nineteen") == "2019"
-
-    assert std("two point five million") == "2500000"
-    assert std("four point two billions") == "4200000000s"
-    assert std("200 thousand") == "200000"
-    assert std("200 thousand dollars") == "$200000"
-    assert std("$20 million") == "$20000000"
-    assert std("€52.4 million") == "€52400000"
-    assert std("£77 thousands") == "£77000s"
-
-    assert std("two double o eight") == "2008"
-
-    assert std("three thousand twenty nine") == "3029"
-    assert std("forty three thousand two hundred sixty") == "43260"
-    assert std("forty three thousand two hundred and sixty") == "43260"
-
-    assert std("nineteen fifties") == "1950s"
-    assert std("thirty first") == "31st"
-    assert std("thirty three thousand and three hundred and thirty third") == "33333rd"
-
-    assert std("three billion") == "3000000000"
-    assert std("millions") == "1000000s"
-
-    assert std("july third twenty twenty") == "july 3rd 2020"
-    assert std("august twenty sixth twenty twenty one") == "august 26th 2021"
-    assert std("3 14") == "3 14"
-    assert std("3.14") == "3.14"
-    assert std("3 point 2") == "3.2"
-    assert std("3 point 14") == "3.14"
-    assert std("fourteen point 4") == "14.4"
-    assert std("two point two five dollars") == "$2.25"
-    assert std("two hundred million dollars") == "$200000000"
-    assert std("$20.1 million") == "$20100000"
-
-    assert std("ninety percent") == "90%"
-    assert std("seventy six per cent") == "76%"
-
-    assert std("double oh seven") == "007"
-    assert std("double zero seven") == "007"
-    assert std("nine one one") == "911"
-    assert std("nine double one") == "911"
-    assert std("one triple oh one") == "10001"
-
-    assert std("two thousandth") == "2000th"
-    assert std("thirty two thousandth") == "32000th"
-
-    assert std("minus 500") == "-500"
-    assert std("positive twenty thousand") == "+20000"
-
-    assert std("two dollars and seventy cents") == "$2.70"
-    assert std("3 cents") == "¢3"
-    assert std("$0.36") == "¢36"
-    assert std("three euros and sixty five cents") == "€3.65"
-
-    assert std("three and a half million") == "3500000"
-    assert std("forty eight and a half dollars") == "$48.5"
-    assert std("b747") == "b 747"
-    assert std("10 th") == "10th"
-    assert std("10th") == "10th"
-
-
-def test_spelling_normalizer():
-    std = EnglishSpellingNormalizer()
-
-    assert std("mobilisation") == "mobilization"
-    assert std("cancelation") == "cancellation"
-
-
-def test_text_normalizer():
-    std = EnglishTextNormalizer()
-    assert std("Let's") == "let us"
-    assert std("he's like") == "he is like"
-    assert std("she's been like") == "she has been like"
-    assert std("10km") == "10 km"
-    assert std("RC232") == "rc 232"
-
-    assert (
-        std("Mr. Park visited Assoc. Prof. Kim Jr.")
-        == "mister park visited associate professor kim junior"
-    )

tests/test_tokenizer.py

@@ -1,14 +0,0 @@
-from whisper.tokenizer import get_tokenizer
-
-
-def test_tokenizer():
-    gpt2_tokenizer = get_tokenizer(multilingual=False)
-    multilingual_tokenizer = get_tokenizer(multilingual=True)
-
-    text = "다람쥐 헌 쳇바퀴에 타고파"
-    gpt2_tokens = gpt2_tokenizer.encode(text)
-    multilingual_tokens = multilingual_tokenizer.encode(text)
-
-    assert gpt2_tokenizer.decode(gpt2_tokens) == text
-    assert multilingual_tokenizer.decode(multilingual_tokens) == text
-    assert len(gpt2_tokens) > len(multilingual_tokens)

tests/test_transcribe.py

@@ -1,20 +0,0 @@
-import os
-
-import pytest
-
-import whisper
-
-
-@pytest.mark.parametrize('model_name', whisper.available_models())
-def test_transcribe(model_name: str):
-    model = whisper.load_model(model_name).cuda()
-    audio_path = os.path.join(os.path.dirname(__file__), "jfk.flac")
-
-    language = "en" if model_name.endswith(".en") else None
-    result = model.transcribe(audio_path, language=language, temperature=0.0)
-    assert result["language"] == "en"
-
-    transcription = result["text"].lower()
-    assert "my fellow americans" in transcription
-    assert "your country" in transcription
-    assert "do for you" in transcription

whisperx/SubtitlesProcessor.py

@@ -0,0 +1,226 @@
+import math
+from whisperx.conjunctions import get_conjunctions, get_comma
+
+def normal_round(n):
+    if n - math.floor(n) < 0.5:
+        return math.floor(n)
+    return math.ceil(n)
+
+
+def format_timestamp(seconds: float, is_vtt: bool = False):
+
+    assert seconds >= 0, "non-negative timestamp expected"
+    milliseconds = round(seconds * 1000.0)
+
+    hours = milliseconds // 3_600_000
+    milliseconds -= hours * 3_600_000
+
+    minutes = milliseconds // 60_000
+    milliseconds -= minutes * 60_000
+
+    seconds = milliseconds // 1_000
+    milliseconds -= seconds * 1_000
+
+    separator = '.' if is_vtt else ','
+    
+    hours_marker = f"{hours:02d}:"
+    return (
+        f"{hours_marker}{minutes:02d}:{seconds:02d}{separator}{milliseconds:03d}"
+    )
+
+
+
+class SubtitlesProcessor:
+    def __init__(self, segments, lang, max_line_length = 45, min_char_length_splitter = 30, is_vtt = False):
+        self.comma = get_comma(lang)
+        self.conjunctions = set(get_conjunctions(lang))
+        self.segments = segments
+        self.lang = lang
+        self.max_line_length = max_line_length
+        self.min_char_length_splitter = min_char_length_splitter
+        self.is_vtt = is_vtt
+        complex_script_languages = ['th', 'lo', 'my', 'km', 'am', 'ko', 'ja', 'zh', 'ti', 'ta', 'te', 'kn', 'ml', 'hi', 'ne', 'mr', 'ar', 'fa', 'ur', 'ka']
+        if self.lang in complex_script_languages:
+            self.max_line_length = 30
+            self.min_char_length_splitter = 20
+
+    def estimate_timestamp_for_word(self, words, i, next_segment_start_time=None):
+        k = 0.25
+        has_prev_end = i > 0 and 'end' in words[i - 1]
+        has_next_start = i < len(words) - 1 and 'start' in words[i + 1]
+
+        if has_prev_end:
+            words[i]['start'] = words[i - 1]['end']
+            if has_next_start:
+                words[i]['end'] = words[i + 1]['start']
+            else:
+                if next_segment_start_time:
+                    words[i]['end'] = next_segment_start_time if next_segment_start_time - words[i - 1]['end'] <= 1 else next_segment_start_time - 0.5
+                else:
+                    words[i]['end'] = words[i]['start'] + len(words[i]['word']) * k
+
+        elif has_next_start:
+            words[i]['start'] = words[i + 1]['start'] - len(words[i]['word']) * k
+            words[i]['end'] = words[i + 1]['start']
+
+        else:
+            if next_segment_start_time:
+                words[i]['start'] = next_segment_start_time - 1
+                words[i]['end'] = next_segment_start_time - 0.5
+            else:
+                words[i]['start'] = 0
+                words[i]['end'] = 0
+
+
+
+    def process_segments(self, advanced_splitting=True):
+        subtitles = []
+        for i, segment in enumerate(self.segments):
+            next_segment_start_time = self.segments[i + 1]['start'] if i + 1 < len(self.segments) else None
+            
+            if advanced_splitting:
+
+                split_points = self.determine_advanced_split_points(segment, next_segment_start_time)
+                subtitles.extend(self.generate_subtitles_from_split_points(segment, split_points, next_segment_start_time))
+            else:
+                words = segment['words']
+                for i, word in enumerate(words):
+                    if 'start' not in word or 'end' not in word:
+                        self.estimate_timestamp_for_word(words, i, next_segment_start_time)
+
+                subtitles.append({
+                'start': segment['start'],
+                'end': segment['end'],
+                'text': segment['text']
+            })
+
+        return subtitles
+
+    def determine_advanced_split_points(self, segment, next_segment_start_time=None):
+        split_points = []
+        last_split_point = 0
+        char_count = 0
+
+        words = segment.get('words', segment['text'].split())
+        add_space = 0 if self.lang in ['zh', 'ja'] else 1
+
+        total_char_count = sum(len(word['word']) if isinstance(word, dict) else len(word) + add_space for word in words)
+        char_count_after = total_char_count
+
+        for i, word in enumerate(words):
+            word_text = word['word'] if isinstance(word, dict) else word
+            word_length = len(word_text) + add_space
+            char_count += word_length
+            char_count_after -= word_length
+
+            char_count_before = char_count - word_length
+
+            if isinstance(word, dict) and ('start' not in word or 'end' not in word):
+                self.estimate_timestamp_for_word(words, i, next_segment_start_time)
+
+            if char_count >= self.max_line_length:
+                midpoint = normal_round((last_split_point + i) / 2)
+                if char_count_before >= self.min_char_length_splitter:
+                    split_points.append(midpoint)
+                    last_split_point = midpoint + 1
+                    char_count = sum(len(words[j]['word']) if isinstance(words[j], dict) else len(words[j]) + add_space for j in range(last_split_point, i + 1))
+
+            elif word_text.endswith(self.comma) and char_count_before >= self.min_char_length_splitter and char_count_after >= self.min_char_length_splitter:
+                split_points.append(i)
+                last_split_point = i + 1
+                char_count = 0
+
+            elif word_text.lower() in self.conjunctions and char_count_before >= self.min_char_length_splitter and char_count_after >= self.min_char_length_splitter:
+                split_points.append(i - 1)
+                last_split_point = i
+                char_count = word_length
+
+        return split_points
+
+    
+    def generate_subtitles_from_split_points(self, segment, split_points, next_start_time=None):
+        subtitles = []
+        
+        words = segment.get('words', segment['text'].split())
+        total_word_count = len(words)
+        total_time = segment['end'] - segment['start']
+        elapsed_time = segment['start']
+        prefix = ' ' if self.lang not in ['zh', 'ja'] else ''
+        start_idx = 0
+        for split_point in split_points:
+
+            fragment_words = words[start_idx:split_point + 1]
+            current_word_count = len(fragment_words)
+            
+
+            if isinstance(fragment_words[0], dict):
+                start_time = fragment_words[0]['start']
+                end_time = fragment_words[-1]['end']
+                next_start_time_for_word = words[split_point + 1]['start'] if split_point + 1 < len(words) else None
+                if next_start_time_for_word and (next_start_time_for_word - end_time) <= 0.8:
+                    end_time = next_start_time_for_word
+            else:
+                fragment = prefix.join(fragment_words).strip()
+                current_duration = (current_word_count / total_word_count) * total_time
+                start_time = elapsed_time
+                end_time = elapsed_time + current_duration
+                elapsed_time += current_duration
+
+
+            subtitles.append({
+                'start': start_time,
+                'end': end_time,
+                'text': fragment if not isinstance(fragment_words[0], dict) else prefix.join(word['word'] for word in fragment_words)
+            })
+            
+            start_idx = split_point + 1
+
+        # Handle the last fragment
+        if start_idx < len(words):
+            fragment_words = words[start_idx:]
+            current_word_count = len(fragment_words)
+            
+            if isinstance(fragment_words[0], dict):
+                start_time = fragment_words[0]['start']
+                end_time = fragment_words[-1]['end']
+            else:
+                fragment = prefix.join(fragment_words).strip()
+                current_duration = (current_word_count / total_word_count) * total_time
+                start_time = elapsed_time
+                end_time = elapsed_time + current_duration
+
+            if next_start_time and (next_start_time - end_time) <= 0.8:
+                end_time = next_start_time
+
+            subtitles.append({
+                'start': start_time,
+                'end': end_time if end_time is not None else segment['end'],
+                'text': fragment if not isinstance(fragment_words[0], dict) else prefix.join(word['word'] for word in fragment_words)
+            })
+            
+        return subtitles
+    
+
+
+    def save(self, filename="subtitles.srt", advanced_splitting=True):
+        
+        subtitles = self.process_segments(advanced_splitting)
+
+        def write_subtitle(file, idx, start_time, end_time, text):
+
+            file.write(f"{idx}\n")
+            file.write(f"{start_time} --> {end_time}\n")
+            file.write(text + "\n\n")
+
+        with open(filename, 'w', encoding='utf-8') as file:
+            if self.is_vtt:
+                file.write("WEBVTT\n\n")
+            
+            if advanced_splitting:
+                for idx, subtitle in enumerate(subtitles, 1):
+                    start_time = format_timestamp(subtitle['start'], self.is_vtt)
+                    end_time = format_timestamp(subtitle['end'], self.is_vtt)
+                    text = subtitle['text'].strip()
+                    write_subtitle(file, idx, start_time, end_time, text)
+
+        return len(subtitles)

whisperx/__init__.py

@@ -1,115 +1,31 @@
-import hashlib
-import io
-import os
-import urllib
-import warnings
-from typing import List, Optional, Union
-
-import torch
-from tqdm import tqdm
-
-from .audio import load_audio, log_mel_spectrogram, pad_or_trim
-from .decoding import DecodingOptions, DecodingResult, decode, detect_language
-from .model import Whisper, ModelDimensions
-from .transcribe import transcribe, transcribe_with_vad, transcribe_with_vad_parallel
-from .alignment import load_align_model, align
-_MODELS = {
-    "tiny.en": "https://openaipublic.azureedge.net/main/whisper/models/d3dd57d32accea0b295c96e26691aa14d8822fac7d9d27d5dc00b4ca2826dd03/tiny.en.pt",
-    "tiny": "https://openaipublic.azureedge.net/main/whisper/models/65147644a518d12f04e32d6f3b26facc3f8dd46e5390956a9424a650c0ce22b9/tiny.pt",
-    "base.en": "https://openaipublic.azureedge.net/main/whisper/models/25a8566e1d0c1e2231d1c762132cd20e0f96a85d16145c3a00adf5d1ac670ead/base.en.pt",
-    "base": "https://openaipublic.azureedge.net/main/whisper/models/ed3a0b6b1c0edf879ad9b11b1af5a0e6ab5db9205f891f668f8b0e6c6326e34e/base.pt",
-    "small.en": "https://openaipublic.azureedge.net/main/whisper/models/f953ad0fd29cacd07d5a9eda5624af0f6bcf2258be67c92b79389873d91e0872/small.en.pt",
-    "small": "https://openaipublic.azureedge.net/main/whisper/models/9ecf779972d90ba49c06d968637d720dd632c55bbf19d441fb42bf17a411e794/small.pt",
-    "medium.en": "https://openaipublic.azureedge.net/main/whisper/models/d7440d1dc186f76616474e0ff0b3b6b879abc9d1a4926b7adfa41db2d497ab4f/medium.en.pt",
-    "medium": "https://openaipublic.azureedge.net/main/whisper/models/345ae4da62f9b3d59415adc60127b97c714f32e89e936602e85993674d08dcb1/medium.pt",
-    "large-v1": "https://openaipublic.azureedge.net/main/whisper/models/e4b87e7e0bf463eb8e6956e646f1e277e901512310def2c24bf0e11bd3c28e9a/large-v1.pt",
-    "large-v2": "https://openaipublic.azureedge.net/main/whisper/models/81f7c96c852ee8fc832187b0132e569d6c3065a3252ed18e56effd0b6a73e524/large-v2.pt",
-    "large": "https://openaipublic.azureedge.net/main/whisper/models/81f7c96c852ee8fc832187b0132e569d6c3065a3252ed18e56effd0b6a73e524/large-v2.pt",
-}
+import importlib
 
 
-def _download(url: str, root: str, in_memory: bool) -> Union[bytes, str]:
-    os.makedirs(root, exist_ok=True)
-
-    expected_sha256 = url.split("/")[-2]
-    download_target = os.path.join(root, os.path.basename(url))
-
-    if os.path.exists(download_target) and not os.path.isfile(download_target):
-        raise RuntimeError(f"{download_target} exists and is not a regular file")
-
-    if os.path.isfile(download_target):
-        with open(download_target, "rb") as f:
-            model_bytes = f.read()
-        if hashlib.sha256(model_bytes).hexdigest() == expected_sha256:
-            return model_bytes if in_memory else download_target
-        else:
-            warnings.warn(f"{download_target} exists, but the SHA256 checksum does not match; re-downloading the file")
-
-    with urllib.request.urlopen(url) as source, open(download_target, "wb") as output:
-        with tqdm(total=int(source.info().get("Content-Length")), ncols=80, unit='iB', unit_scale=True, unit_divisor=1024) as loop:
-            while True:
-                buffer = source.read(8192)
-                if not buffer:
-                    break
-
-                output.write(buffer)
-                loop.update(len(buffer))
-
-    model_bytes = open(download_target, "rb").read()
-    if hashlib.sha256(model_bytes).hexdigest() != expected_sha256:
-        raise RuntimeError("Model has been downloaded but the SHA256 checksum does not not match. Please retry loading the model.")
-
-    return model_bytes if in_memory else download_target
+def _lazy_import(name):
+    module = importlib.import_module(f"whisperx.{name}")
+    return module
 
 
-def available_models() -> List[str]:
-    """Returns the names of available models"""
-    return list(_MODELS.keys())
+def load_align_model(*args, **kwargs):
+    alignment = _lazy_import("alignment")
+    return alignment.load_align_model(*args, **kwargs)
 
 
-def load_model(name: str, device: Optional[Union[str, torch.device]] = None, download_root: str = None, in_memory: bool = False) -> Whisper:
-    """
-    Load a Whisper ASR model
+def align(*args, **kwargs):
+    alignment = _lazy_import("alignment")
+    return alignment.align(*args, **kwargs)
 
-    Parameters
-    ----------
-    name : str
-        one of the official model names listed by `whisper.available_models()`, or
-        path to a model checkpoint containing the model dimensions and the model state_dict.
-    device : Union[str, torch.device]
-        the PyTorch device to put the model into
-    download_root: str
-        path to download the model files; by default, it uses "~/.cache/whisper"
-    in_memory: bool
-        whether to preload the model weights into host memory
 
-    Returns
-    -------
-    model : Whisper
-        The Whisper ASR model instance
-    """
+def load_model(*args, **kwargs):
+    asr = _lazy_import("asr")
+    return asr.load_model(*args, **kwargs)
 
-    if device is None:
-        device = "cuda" if torch.cuda.is_available() else "cpu"
-    if download_root is None:
-        download_root = os.getenv(
-            "XDG_CACHE_HOME", 
-            os.path.join(os.path.expanduser("~"), ".cache", "whisper")
-        )
 
-    if name in _MODELS:
-        checkpoint_file = _download(_MODELS[name], download_root, in_memory)
-    elif os.path.isfile(name):
-        checkpoint_file = open(name, "rb").read() if in_memory else name
-    else:
-        raise RuntimeError(f"Model {name} not found; available models = {available_models()}")
+def load_audio(*args, **kwargs):
+    audio = _lazy_import("audio")
+    return audio.load_audio(*args, **kwargs)
 
-    with (io.BytesIO(checkpoint_file) if in_memory else open(checkpoint_file, "rb")) as fp:
-        checkpoint = torch.load(fp, map_location=device)
-    del checkpoint_file
 
-    dims = ModelDimensions(**checkpoint["dims"])
-    model = Whisper(dims)
-    model.load_state_dict(checkpoint["model_state_dict"])
-
-    return model.to(device)
+def assign_word_speakers(*args, **kwargs):
+    diarize = _lazy_import("diarize")
+    return diarize.assign_word_speakers(*args, **kwargs)

whisperx/__main__.py

@@ -1,4 +1,88 @@
-from .transcribe import cli
+import argparse
+import importlib.metadata
+import platform
+
+import torch
+
+from whisperx.utils import (LANGUAGES, TO_LANGUAGE_CODE, optional_float,
+                            optional_int, str2bool)
 
 
-cli()
+def cli():
+    # fmt: off
+    parser = argparse.ArgumentParser(formatter_class=argparse.ArgumentDefaultsHelpFormatter)
+    parser.add_argument("audio", nargs="+", type=str, help="audio file(s) to transcribe")
+    parser.add_argument("--model", default="small", help="name of the Whisper model to use")
+    parser.add_argument("--model_cache_only", type=str2bool, default=False, help="If True, will not attempt to download models, instead using cached models from --model_dir")
+    parser.add_argument("--model_dir", type=str, default=None, help="the path to save model files; uses ~/.cache/whisper by default")
+    parser.add_argument("--device", default="cuda" if torch.cuda.is_available() else "cpu", help="device to use for PyTorch inference")
+    parser.add_argument("--device_index", default=0, type=int, help="device index to use for FasterWhisper inference")
+    parser.add_argument("--batch_size", default=8, type=int, help="the preferred batch size for inference")
+    parser.add_argument("--compute_type", default="float16", type=str, choices=["float16", "float32", "int8"], help="compute type for computation")
+
+    parser.add_argument("--output_dir", "-o", type=str, default=".", help="directory to save the outputs")
+    parser.add_argument("--output_format", "-f", type=str, default="all", choices=["all", "srt", "vtt", "txt", "tsv", "json", "aud"], help="format of the output file; if not specified, all available formats will be produced")
+    parser.add_argument("--verbose", type=str2bool, default=True, help="whether to print out the progress and debug messages")
+
+    parser.add_argument("--task", type=str, default="transcribe", choices=["transcribe", "translate"], help="whether to perform X->X speech recognition ('transcribe') or X->English translation ('translate')")
+    parser.add_argument("--language", type=str, default=None, choices=sorted(LANGUAGES.keys()) + sorted([k.title() for k in TO_LANGUAGE_CODE.keys()]), help="language spoken in the audio, specify None to perform language detection")
+
+    # alignment params
+    parser.add_argument("--align_model", default=None, help="Name of phoneme-level ASR model to do alignment")
+    parser.add_argument("--interpolate_method", default="nearest", choices=["nearest", "linear", "ignore"], help="For word .srt, method to assign timestamps to non-aligned words, or merge them into neighbouring.")
+    parser.add_argument("--no_align", action='store_true', help="Do not perform phoneme alignment")
+    parser.add_argument("--return_char_alignments", action='store_true', help="Return character-level alignments in the output json file")
+
+    # vad params
+    parser.add_argument("--vad_method", type=str, default="pyannote", choices=["pyannote", "silero"], help="VAD method to be used")
+    parser.add_argument("--vad_onset", type=float, default=0.500, help="Onset threshold for VAD (see pyannote.audio), reduce this if speech is not being detected")
+    parser.add_argument("--vad_offset", type=float, default=0.363, help="Offset threshold for VAD (see pyannote.audio), reduce this if speech is not being detected.")
+    parser.add_argument("--chunk_size", type=int, default=30, help="Chunk size for merging VAD segments. Default is 30, reduce this if the chunk is too long.")
+
+    # diarization params
+    parser.add_argument("--diarize", action="store_true", help="Apply diarization to assign speaker labels to each segment/word")
+    parser.add_argument("--min_speakers", default=None, type=int, help="Minimum number of speakers to in audio file")
+    parser.add_argument("--max_speakers", default=None, type=int, help="Maximum number of speakers to in audio file")
+    parser.add_argument("--diarize_model", default="pyannote/speaker-diarization-3.1", type=str, help="Name of the speaker diarization model to use")
+
+    parser.add_argument("--temperature", type=float, default=0, help="temperature to use for sampling")
+    parser.add_argument("--best_of", type=optional_int, default=5, help="number of candidates when sampling with non-zero temperature")
+    parser.add_argument("--beam_size", type=optional_int, default=5, help="number of beams in beam search, only applicable when temperature is zero")
+    parser.add_argument("--patience", type=float, default=1.0, help="optional patience value to use in beam decoding, as in https://arxiv.org/abs/2204.05424, the default (1.0) is equivalent to conventional beam search")
+    parser.add_argument("--length_penalty", type=float, default=1.0, help="optional token length penalty coefficient (alpha) as in https://arxiv.org/abs/1609.08144, uses simple length normalization by default")
+
+    parser.add_argument("--suppress_tokens", type=str, default="-1", help="comma-separated list of token ids to suppress during sampling; '-1' will suppress most special characters except common punctuations")
+    parser.add_argument("--suppress_numerals", action="store_true", help="whether to suppress numeric symbols and currency symbols during sampling, since wav2vec2 cannot align them correctly")
+
+    parser.add_argument("--initial_prompt", type=str, default=None, help="optional text to provide as a prompt for the first window.")
+    parser.add_argument("--condition_on_previous_text", type=str2bool, default=False, help="if True, provide the previous output of the model as a prompt for the next window; disabling may make the text inconsistent across windows, but the model becomes less prone to getting stuck in a failure loop")
+    parser.add_argument("--fp16", type=str2bool, default=True, help="whether to perform inference in fp16; True by default")
+
+    parser.add_argument("--temperature_increment_on_fallback", type=optional_float, default=0.2, help="temperature to increase when falling back when the decoding fails to meet either of the thresholds below")
+    parser.add_argument("--compression_ratio_threshold", type=optional_float, default=2.4, help="if the gzip compression ratio is higher than this value, treat the decoding as failed")
+    parser.add_argument("--logprob_threshold", type=optional_float, default=-1.0, help="if the average log probability is lower than this value, treat the decoding as failed")
+    parser.add_argument("--no_speech_threshold", type=optional_float, default=0.6, help="if the probability of the <|nospeech|> token is higher than this value AND the decoding has failed due to `logprob_threshold`, consider the segment as silence")
+
+    parser.add_argument("--max_line_width", type=optional_int, default=None, help="(not possible with --no_align) the maximum number of characters in a line before breaking the line")
+    parser.add_argument("--max_line_count", type=optional_int, default=None, help="(not possible with --no_align) the maximum number of lines in a segment")
+    parser.add_argument("--highlight_words", type=str2bool, default=False, help="(not possible with --no_align) underline each word as it is spoken in srt and vtt")
+    parser.add_argument("--segment_resolution", type=str, default="sentence", choices=["sentence", "chunk"], help="(not possible with --no_align) the maximum number of characters in a line before breaking the line")
+
+    parser.add_argument("--threads", type=optional_int, default=0, help="number of threads used by torch for CPU inference; supercedes MKL_NUM_THREADS/OMP_NUM_THREADS")
+
+    parser.add_argument("--hf_token", type=str, default=None, help="Hugging Face Access Token to access PyAnnote gated models")
+
+    parser.add_argument("--print_progress", type=str2bool, default = False, help = "if True, progress will be printed in transcribe() and align() methods.")
+    parser.add_argument("--version", "-V", action="version", version=f"%(prog)s {importlib.metadata.version('whisperx')}",help="Show whisperx version information and exit")
+    parser.add_argument("--python-version", "-P", action="version", version=f"Python {platform.python_version()} ({platform.python_implementation()})",help="Show python version information and exit")
+    # fmt: on
+
+    args = parser.parse_args().__dict__
+
+    from whisperx.transcribe import transcribe_task
+
+    transcribe_task(args, parser)
+
+
+if __name__ == "__main__":
+    cli()

whisperx/alignment.py

@@ -1,17 +1,30 @@
-""""
+"""
 Forced Alignment with Whisper
 C. Max Bain
 """
+import math
+
+from dataclasses import dataclass
+from typing import Iterable, Optional, Union, List
+
 import numpy as np
 import pandas as pd
-from typing import List, Union, Iterator, TYPE_CHECKING
-from transformers import Wav2Vec2ForCTC, Wav2Vec2Processor
-import torchaudio
 import torch
-from dataclasses import dataclass
-from .audio import SAMPLE_RATE, load_audio
-from .utils import interpolate_nans
+import torchaudio
+from transformers import Wav2Vec2ForCTC, Wav2Vec2Processor
 
+from whisperx.audio import SAMPLE_RATE, load_audio
+from whisperx.utils import interpolate_nans
+from whisperx.types import (
+    AlignedTranscriptionResult,
+    SingleSegment,
+    SingleAlignedSegment,
+    SingleWordSegment,
+    SegmentData,
+)
+from nltk.tokenize.punkt import PunktSentenceTokenizer, PunktParameters
+
+PUNKT_ABBREVIATIONS = ['dr', 'vs', 'mr', 'mrs', 'prof']
 
 LANGUAGES_WITHOUT_SPACES = ["ja", "zh"]
 
@@ -30,6 +43,7 @@ DEFAULT_ALIGN_MODELS_HF = {
     "uk": "Yehor/wav2vec2-xls-r-300m-uk-with-small-lm",
     "pt": "jonatasgrosman/wav2vec2-large-xlsr-53-portuguese",
     "ar": "jonatasgrosman/wav2vec2-large-xlsr-53-arabic",
+    "cs": "comodoro/wav2vec2-xls-r-300m-cs-250",
     "ru": "jonatasgrosman/wav2vec2-large-xlsr-53-russian",
     "pl": "jonatasgrosman/wav2vec2-large-xlsr-53-polish",
     "hu": "jonatasgrosman/wav2vec2-large-xlsr-53-hungarian",
@@ -37,10 +51,30 @@ DEFAULT_ALIGN_MODELS_HF = {
     "fa": "jonatasgrosman/wav2vec2-large-xlsr-53-persian",
     "el": "jonatasgrosman/wav2vec2-large-xlsr-53-greek",
     "tr": "mpoyraz/wav2vec2-xls-r-300m-cv7-turkish",
+    "da": "saattrupdan/wav2vec2-xls-r-300m-ftspeech",
+    "he": "imvladikon/wav2vec2-xls-r-300m-hebrew",
+    "vi": 'nguyenvulebinh/wav2vec2-base-vi',
+    "ko": "kresnik/wav2vec2-large-xlsr-korean",
+    "ur": "kingabzpro/wav2vec2-large-xls-r-300m-Urdu",
+    "te": "anuragshas/wav2vec2-large-xlsr-53-telugu",
+    "hi": "theainerd/Wav2Vec2-large-xlsr-hindi",
+    "ca": "softcatala/wav2vec2-large-xlsr-catala",
+    "ml": "gvs/wav2vec2-large-xlsr-malayalam",
+    "no": "NbAiLab/nb-wav2vec2-1b-bokmaal-v2",
+    "nn": "NbAiLab/nb-wav2vec2-1b-nynorsk",
+    "sk": "comodoro/wav2vec2-xls-r-300m-sk-cv8",
+    "sl": "anton-l/wav2vec2-large-xlsr-53-slovenian",
+    "hr": "classla/wav2vec2-xls-r-parlaspeech-hr",
+    "ro": "gigant/romanian-wav2vec2",
+    "eu": "stefan-it/wav2vec2-large-xlsr-53-basque",
+    "gl": "ifrz/wav2vec2-large-xlsr-galician",
+    "ka": "xsway/wav2vec2-large-xlsr-georgian",
+    "lv": "jimregan/wav2vec2-large-xlsr-latvian-cv",
+    "tl": "Khalsuu/filipino-wav2vec2-l-xls-r-300m-official",
 }
 
 
-def load_align_model(language_code, device, model_name=None):
+def load_align_model(language_code: str, device: str, model_name: Optional[str] = None, model_dir=None):
     if model_name is None:
         # use default model
         if language_code in DEFAULT_ALIGN_MODELS_TORCH:
@@ -55,13 +89,13 @@ def load_align_model(language_code, device, model_name=None):
     if model_name in torchaudio.pipelines.__all__:
         pipeline_type = "torchaudio"
         bundle = torchaudio.pipelines.__dict__[model_name]
-        align_model = bundle.get_model().to(device)
+        align_model = bundle.get_model(dl_kwargs={"model_dir": model_dir}).to(device)
         labels = bundle.get_labels()
         align_dictionary = {c.lower(): i for i, c in enumerate(labels)}
     else:
         try:
-            processor = Wav2Vec2Processor.from_pretrained(model_name)
-            align_model = Wav2Vec2ForCTC.from_pretrained(model_name)
+            processor = Wav2Vec2Processor.from_pretrained(model_name, cache_dir=model_dir)
+            align_model = Wav2Vec2ForCTC.from_pretrained(model_name, cache_dir=model_dir)
         except Exception as e:
             print(e)
             print(f"Error loading model from huggingface, check https://huggingface.co/models for finetuned wav2vec2.0 models")
@@ -77,431 +111,474 @@ def load_align_model(language_code, device, model_name=None):
 
 
 def align(
-    transcript: Iterator[dict],
+    transcript: Iterable[SingleSegment],
     model: torch.nn.Module,
     align_model_metadata: dict,
     audio: Union[str, np.ndarray, torch.Tensor],
     device: str,
-    extend_duration: float = 0.0,
-    start_from_previous: bool = True,
     interpolate_method: str = "nearest",
-):
+    return_char_alignments: bool = False,
+    print_progress: bool = False,
+    combined_progress: bool = False,
+) -> AlignedTranscriptionResult:
+    """
+    Align phoneme recognition predictions to known transcription.
     """
-    Force align phoneme recognition predictions to known transcription
-
-    Parameters
-    ----------
-    transcript: Iterator[dict]
-        The Whisper model instance
     
-    model: torch.nn.Module
-        Alignment model (wav2vec2)
-
-    audio: Union[str, np.ndarray, torch.Tensor]
-        The path to the audio file to open, or the audio waveform
-
-    device: str
-        cuda device
-
-    diarization: pd.DataFrame {'start': List[float], 'end': List[float], 'speaker': List[float]}
-        diarization segments with speaker labels.
-
-    extend_duration: float
-        Amount to pad input segments by. If not using vad--filter then recommended to use 2 seconds
-
-        If the gzip compression ratio is above this value, treat as failed
-
-    interpolate_method: str ["nearest", "linear", "ignore"]
-        Method to assign timestamps to non-aligned words. Words are not able to be aligned when none of the characters occur in the align model dictionary.
-        "nearest" copies timestamp of nearest word within the segment. "linear" is linear interpolation. "drop" removes that word from output.
-
-    Returns
-    -------
-    A dictionary containing the resulting text ("text") and segment-level details ("segments"), and
-    the spoken language ("language"), which is detected when `decode_options["language"]` is None.
-    """
     if not torch.is_tensor(audio):
         if isinstance(audio, str):
             audio = load_audio(audio)
         audio = torch.from_numpy(audio)
     if len(audio.shape) == 1:
         audio = audio.unsqueeze(0)
-
+    
     MAX_DURATION = audio.shape[1] / SAMPLE_RATE
 
     model_dictionary = align_model_metadata["dictionary"]
     model_lang = align_model_metadata["language"]
     model_type = align_model_metadata["type"]
 
-    aligned_segments = []
-
-    prev_t2 = 0
-
-    char_segments_arr = {
-        "segment-idx": [],
-        "subsegment-idx": [],
-        "word-idx": [],
-        "char": [],
-        "start": [],
-        "end": [],
-        "score": [],
-    }
-
+    # 1. Preprocess to keep only characters in dictionary
+    total_segments = len(transcript)
+    # Store temporary processing values
+    segment_data: dict[int, SegmentData] = {}
     for sdx, segment in enumerate(transcript):
-        while True:
-            segment_align_success = False
-
-            # strip spaces at beginning / end, but keep track of the amount.
-            num_leading = len(segment["text"]) - len(segment["text"].lstrip())
-            num_trailing = len(segment["text"]) - len(segment["text"].rstrip())
-            transcription = segment["text"]
-
-            # TODO: convert number tokenizer / symbols to phonetic words for alignment.
-            # e.g. "$300" -> "three hundred dollars"
-            # currently "$300" is ignored since no characters present in the phonetic dictionary
-
-            # split into words
-            if model_lang not in LANGUAGES_WITHOUT_SPACES:
-                per_word = transcription.split(" ")
-            else:
-                per_word = transcription
-
-            # first check that characters in transcription can be aligned (they are contained in align model"s dictionary)
-            clean_char, clean_cdx = [], []
-            for cdx, char in enumerate(transcription):
-                char_ = char.lower()
-                # wav2vec2 models use "|" character to represent spaces
-                if model_lang not in LANGUAGES_WITHOUT_SPACES:
-                    char_ = char_.replace(" ", "|")
-                
-                # ignore whitespace at beginning and end of transcript
-                if cdx < num_leading:
-                    pass
-                elif cdx > len(transcription) - num_trailing - 1:
-                    pass
-                elif char_ in model_dictionary.keys():
-                    clean_char.append(char_)
-                    clean_cdx.append(cdx)
-
-            clean_wdx = []
-            for wdx, wrd in enumerate(per_word):
-                if any([c in model_dictionary.keys() for c in wrd]):
-                    clean_wdx.append(wdx)
-
-            # if no characters are in the dictionary, then we skip this segment...
-            if len(clean_char) == 0:
-                print(f'Failed to align segment ("{segment["text"]}"): no characters in this segment found in model dictionary, resorting to original...')
-                break          
-           
-            transcription_cleaned = "".join(clean_char)
-            tokens = [model_dictionary[c] for c in transcription_cleaned]
-
-            # we only pad if not using VAD filtering
-            if "seg_text" not in segment:
-                # pad according original timestamps
-                t1 = max(segment["start"] - extend_duration, 0)
-                t2 = min(segment["end"] + extend_duration, MAX_DURATION)
-
-            # use prev_t2 as current t1 if it"s later
-            if start_from_previous and t1 < prev_t2:
-                t1 = prev_t2
-
-            # check if timestamp range is still valid
-            if t1 >= MAX_DURATION:
-                print("Failed to align segment: original start time longer than audio duration, skipping...")
-                break
-            if t2 - t1 < 0.02:
-                print("Failed to align segment: duration smaller than 0.02s time precision")
-                break
-
-            f1 = int(t1 * SAMPLE_RATE)
-            f2 = int(t2 * SAMPLE_RATE)
-
-            waveform_segment = audio[:, f1:f2]
-
-            with torch.inference_mode():
-                if model_type == "torchaudio":
-                    emissions, _ = model(waveform_segment.to(device))
-                elif model_type == "huggingface":
-                    emissions = model(waveform_segment.to(device)).logits
-                else:
-                    raise NotImplementedError(f"Align model of type {model_type} not supported.")
-                emissions = torch.log_softmax(emissions, dim=-1)
-
-            emission = emissions[0].cpu().detach()
-
-            trellis = get_trellis(emission, tokens)
-            path = backtrack(trellis, emission, tokens)
-            if path is None:
-                print(f'Failed to align segment ("{segment["text"]}"): backtrack failed, resorting to original...')
-                break
-            char_segments = merge_repeats(path, transcription_cleaned)
-            # word_segments = merge_words(char_segments)
+        # strip spaces at beginning / end, but keep track of the amount.
+        if print_progress:
+            base_progress = ((sdx + 1) / total_segments) * 100
+            percent_complete = (50 + base_progress / 2) if combined_progress else base_progress
+            print(f"Progress: {percent_complete:.2f}%...")
             
+        num_leading = len(segment["text"]) - len(segment["text"].lstrip())
+        num_trailing = len(segment["text"]) - len(segment["text"].rstrip())
+        text = segment["text"]
 
-            # sub-segments
-            if "seg-text" not in segment:
-                segment["seg-text"] = [transcription]
-                
-            v = 0
-            seg_lens = [0] + [len(x) for x in segment["seg-text"]]
-            seg_lens_cumsum = [v := v + n for n in seg_lens]
-            sub_seg_idx = 0
+        # split into words
+        if model_lang not in LANGUAGES_WITHOUT_SPACES:
+            per_word = text.split(" ")
+        else:
+            per_word = text
 
-            wdx = 0
-            duration = t2 - t1
-            ratio = duration * waveform_segment.size(0) / (trellis.size(0) - 1)
-            for cdx, char in enumerate(transcription + " "):
-                is_last = False
-                if cdx == len(transcription):
-                    break
-                elif cdx+1 == len(transcription):
-                    is_last = True
+        clean_char, clean_cdx = [], []
+        for cdx, char in enumerate(text):
+            char_ = char.lower()
+            # wav2vec2 models use "|" character to represent spaces
+            if model_lang not in LANGUAGES_WITHOUT_SPACES:
+                char_ = char_.replace(" ", "|")
+            
+            # ignore whitespace at beginning and end of transcript
+            if cdx < num_leading:
+                pass
+            elif cdx > len(text) - num_trailing - 1:
+                pass
+            elif char_ in model_dictionary.keys():
+                clean_char.append(char_)
+                clean_cdx.append(cdx)
+            else:
+                # add placeholder
+                clean_char.append('*')
+                clean_cdx.append(cdx)
+
+        clean_wdx = []
+        for wdx, wrd in enumerate(per_word):
+            if any([c in model_dictionary.keys() for c in wrd.lower()]):
+                clean_wdx.append(wdx)
+            else:
+                # index for placeholder
+                clean_wdx.append(wdx)
 
                 
-                start, end, score = None, None, None
-                if cdx in clean_cdx:
-                    char_seg = char_segments[clean_cdx.index(cdx)]
-                    start = char_seg.start * ratio + t1
-                    end = char_seg.end * ratio + t1
-                    score = char_seg.score        
-
-                char_segments_arr["char"].append(char)
-                char_segments_arr["start"].append(start)
-                char_segments_arr["end"].append(end)
-                char_segments_arr["score"].append(score)
-                char_segments_arr["word-idx"].append(wdx)
-                char_segments_arr["segment-idx"].append(sdx)
-                char_segments_arr["subsegment-idx"].append(sub_seg_idx)
-
-                # word-level info
-                if model_lang in LANGUAGES_WITHOUT_SPACES:
-                    # character == word
-                    wdx += 1
-                elif is_last or transcription[cdx+1] == " " or cdx == seg_lens_cumsum[sub_seg_idx+1] - 1:
-                    wdx += 1
-
-                if is_last or cdx == seg_lens_cumsum[sub_seg_idx+1] - 1:
-                    wdx = 0
-                    sub_seg_idx += 1
-
-            prev_t2 = segment["end"]
-
-            segment_align_success = True
-            # end while True loop
-            break
-
-        # reset prev_t2 due to drifting issues
-        if not segment_align_success:
-            prev_t2 = 0
-        
-    char_segments_arr = pd.DataFrame(char_segments_arr)
-    not_space = char_segments_arr["char"] != " "
-
-    per_seg_grp = char_segments_arr.groupby(["segment-idx", "subsegment-idx"], as_index = False)
-    char_segments_arr = per_seg_grp.apply(lambda x: x.reset_index(drop = True)).reset_index()
-    per_word_grp = char_segments_arr[not_space].groupby(["segment-idx", "subsegment-idx", "word-idx"])
-    per_subseg_grp = char_segments_arr[not_space].groupby(["segment-idx", "subsegment-idx"])
-    per_seg_grp = char_segments_arr[not_space].groupby(["segment-idx"])
-    char_segments_arr["local-char-idx"] = char_segments_arr.groupby(["segment-idx", "subsegment-idx"]).cumcount()
-    per_word_grp = char_segments_arr[not_space].groupby(["segment-idx", "subsegment-idx", "word-idx"]) # regroup
-
-    word_segments_arr = {}
-
-    # start of word is first char with a timestamp
-    word_segments_arr["start"] = per_word_grp["start"].min().values
-    # end of word is last char with a timestamp
-    word_segments_arr["end"] = per_word_grp["end"].max().values
-    # score of word is mean (excluding nan)
-    word_segments_arr["score"] = per_word_grp["score"].mean().values
-
-    word_segments_arr["segment-text-start"] = per_word_grp["local-char-idx"].min().astype(int).values
-    word_segments_arr["segment-text-end"] = per_word_grp["local-char-idx"].max().astype(int).values+1
-    word_segments_arr = pd.DataFrame(word_segments_arr)
-
-    word_segments_arr[["segment-idx", "subsegment-idx", "word-idx"]] = per_word_grp["local-char-idx"].min().reset_index()[["segment-idx", "subsegment-idx", "word-idx"]].astype(int)
-    segments_arr = {}
-    segments_arr["start"] = per_subseg_grp["start"].min().reset_index()["start"]
-    segments_arr["end"] = per_subseg_grp["end"].max().reset_index()["end"]
-    segments_arr = pd.DataFrame(segments_arr)
-    segments_arr[["segment-idx", "subsegment-idx-start"]] = per_subseg_grp["start"].min().reset_index()[["segment-idx", "subsegment-idx"]]
-    segments_arr["subsegment-idx-end"] = segments_arr["subsegment-idx-start"] + 1
-
-    # interpolate missing words / sub-segments
-    if interpolate_method != "ignore":
-        wrd_subseg_grp = word_segments_arr.groupby(["segment-idx", "subsegment-idx"], group_keys=False)
-        wrd_seg_grp = word_segments_arr.groupby(["segment-idx"], group_keys=False)
-        # we still know which word timestamps are interpolated because their score == nan
-        word_segments_arr["start"] = wrd_subseg_grp['start'].apply(lambda group: interpolate_nans(group, method=interpolate_method))
-        word_segments_arr["end"] = wrd_subseg_grp['end'].apply(lambda group: interpolate_nans(group, method=interpolate_method))
-
-        word_segments_arr["start"] = wrd_seg_grp['start'].apply(lambda group: interpolate_nans(group, method=interpolate_method))
-        word_segments_arr["end"] = wrd_seg_grp['end'].apply(lambda group: interpolate_nans(group, method=interpolate_method))
-
-        sub_seg_grp =  segments_arr.groupby(["segment-idx"], group_keys=False)
-        segments_arr['start'] = sub_seg_grp['start'].apply(lambda group: interpolate_nans(group, method=interpolate_method))
-        segments_arr['end'] = sub_seg_grp['end'].apply(lambda group: interpolate_nans(group, method=interpolate_method))
-
-        # merge words & subsegments which are missing times
-        word_grp = word_segments_arr.groupby(["segment-idx", "subsegment-idx", "end"])
-
-        word_segments_arr["segment-text-start"] = word_grp["segment-text-start"].transform(min)
-        word_segments_arr["segment-text-end"] = word_grp["segment-text-end"].transform(max)
-        word_segments_arr.drop_duplicates(subset=["segment-idx", "subsegment-idx", "end"], inplace=True)
-
-        seg_grp_dup = segments_arr.groupby(["segment-idx", "start", "end"])
-        segments_arr["subsegment-idx-start"] = seg_grp_dup["subsegment-idx-start"].transform(min)
-        segments_arr["subsegment-idx-end"] = seg_grp_dup["subsegment-idx-end"].transform(max)
-        segments_arr.drop_duplicates(subset=["segment-idx", "subsegment-idx-start", "subsegment-idx-end"], inplace=True)
-    else:
-        word_segments_arr.dropna(inplace=True)
-        segments_arr.dropna(inplace=True)
-
-    # if some segments still have missing timestamps (usually because all numerals / symbols), then use original timestamps...
-    segments_arr['start'].fillna(pd.Series([x['start'] for x in transcript]), inplace=True)
-    segments_arr['end'].fillna(pd.Series([x['end'] for x in transcript]), inplace=True)
-    segments_arr['subsegment-idx-start'].fillna(0, inplace=True)
-    segments_arr['subsegment-idx-end'].fillna(1, inplace=True)
-
-
-    aligned_segments = []
-    aligned_segments_word = []
-
-    word_segments_arr.set_index(["segment-idx", "subsegment-idx"], inplace=True)
-    char_segments_arr.set_index(["segment-idx", "subsegment-idx", "word-idx"], inplace=True)
-
-    for sdx, srow in segments_arr.iterrows():
-
-        seg_idx = int(srow["segment-idx"])
-        try:
-            sub_start = int(srow["subsegment-idx-start"])
-        except:
-            import pdb; pdb.set_trace()
-        sub_end = int(srow["subsegment-idx-end"])
-
-        seg = transcript[seg_idx]
-        text = "".join(seg["seg-text"][sub_start:sub_end])
-
-        wseg = word_segments_arr.loc[seg_idx].loc[sub_start:sub_end-1]
-        wseg["start"].fillna(srow["start"], inplace=True)
-        wseg["end"].fillna(srow["end"], inplace=True)
-        wseg["segment-text-start"].fillna(0, inplace=True)
-        wseg["segment-text-end"].fillna(len(text)-1, inplace=True)
-
-        cseg = char_segments_arr.loc[seg_idx].loc[sub_start:sub_end-1]
-        # fixes bug for single segment in transcript
-        cseg['segment-text-start'] = cseg['level_1'] if 'level_1' in cseg else 0
-        cseg['segment-text-end'] = cseg['level_1'] + 1 if 'level_1' in cseg else 1
-        if 'level_1' in cseg: del cseg['level_1']
-        if 'level_0' in cseg: del cseg['level_0']
-        cseg.reset_index(inplace=True)
-        aligned_segments.append(
-            {
-                "start": srow["start"],
-                "end": srow["end"],
-                "text": text,
-                "word-segments": wseg,
-                "char-segments": cseg
-            }
-        )
-
-        def get_raw_text(word_row):
-            return seg["seg-text"][word_row.name][int(word_row["segment-text-start"]):int(word_row["segment-text-end"])+1]
-
-        wdx = 0
-        curr_text = get_raw_text(wseg.iloc[wdx])
-        if len(wseg) > 1:
-            for _, wrow in wseg.iloc[1:].iterrows():
-                if wrow['start'] != wseg.iloc[wdx]['start']:
-                    aligned_segments_word.append(
-                        {
-                            "text": curr_text.strip(),
-                            "start": wseg.iloc[wdx]["start"],
-                            "end": wseg.iloc[wdx]["end"],
-                        }
-                    )
-                    curr_text = ""
-                curr_text += " " + get_raw_text(wrow)
-                wdx += 1
-        aligned_segments_word.append(
-            {
-                "text": curr_text.strip(),
-                "start": wseg.iloc[wdx]["start"],
-                "end": wseg.iloc[wdx]["end"]
-            }
-        )
+        punkt_param = PunktParameters()
+        punkt_param.abbrev_types = set(PUNKT_ABBREVIATIONS)
+        sentence_splitter = PunktSentenceTokenizer(punkt_param)
+        sentence_spans = list(sentence_splitter.span_tokenize(text))
 
+        segment_data[sdx] = {
+            "clean_char": clean_char,
+            "clean_cdx": clean_cdx,
+            "clean_wdx": clean_wdx,
+            "sentence_spans": sentence_spans
+        }
+            
+    aligned_segments: List[SingleAlignedSegment] = []
     
-    return {"segments": aligned_segments, "word_segments": aligned_segments_word}
+    # 2. Get prediction matrix from alignment model & align
+    for sdx, segment in enumerate(transcript):
+        
+        t1 = segment["start"]
+        t2 = segment["end"]
+        text = segment["text"]
 
+        aligned_seg: SingleAlignedSegment = {
+            "start": t1,
+            "end": t2,
+            "text": text,
+            "words": [],
+            "chars": None,
+        }
+
+        if return_char_alignments:
+            aligned_seg["chars"] = []
+
+        # check we can align
+        if len(segment_data[sdx]["clean_char"]) == 0:
+            print(f'Failed to align segment ("{segment["text"]}"): no characters in this segment found in model dictionary, resorting to original...')
+            aligned_segments.append(aligned_seg)
+            continue
+
+        if t1 >= MAX_DURATION:
+            print(f'Failed to align segment ("{segment["text"]}"): original start time longer than audio duration, skipping...')
+            aligned_segments.append(aligned_seg)
+            continue
+
+        text_clean = "".join(segment_data[sdx]["clean_char"])
+        tokens = [model_dictionary.get(c, -1) for c in text_clean]
+
+        f1 = int(t1 * SAMPLE_RATE)
+        f2 = int(t2 * SAMPLE_RATE)
+
+        # TODO: Probably can get some speedup gain with batched inference here
+        waveform_segment = audio[:, f1:f2]
+        # Handle the minimum input length for wav2vec2 models
+        if waveform_segment.shape[-1] < 400:
+            lengths = torch.as_tensor([waveform_segment.shape[-1]]).to(device)
+            waveform_segment = torch.nn.functional.pad(
+                waveform_segment, (0, 400 - waveform_segment.shape[-1])
+            )
+        else:
+            lengths = None
+            
+        with torch.inference_mode():
+            if model_type == "torchaudio":
+                emissions, _ = model(waveform_segment.to(device), lengths=lengths)
+            elif model_type == "huggingface":
+                emissions = model(waveform_segment.to(device)).logits
+            else:
+                raise NotImplementedError(f"Align model of type {model_type} not supported.")
+            emissions = torch.log_softmax(emissions, dim=-1)
+
+        emission = emissions[0].cpu().detach()
+
+        blank_id = 0
+        for char, code in model_dictionary.items():
+            if char == '[pad]' or char == '<pad>':
+                blank_id = code
+
+        trellis = get_trellis(emission, tokens, blank_id)
+        # path = backtrack(trellis, emission, tokens, blank_id)
+        path = backtrack_beam(trellis, emission, tokens, blank_id, beam_width=2)
+
+        if path is None:
+            print(f'Failed to align segment ("{segment["text"]}"): backtrack failed, resorting to original...')
+            aligned_segments.append(aligned_seg)
+            continue
+
+        char_segments = merge_repeats(path, text_clean)
+
+        duration = t2 - t1
+        ratio = duration * waveform_segment.size(0) / (trellis.size(0) - 1)
+
+        # assign timestamps to aligned characters
+        char_segments_arr = []
+        word_idx = 0
+        for cdx, char in enumerate(text):
+            start, end, score = None, None, None
+            if cdx in segment_data[sdx]["clean_cdx"]:
+                char_seg = char_segments[segment_data[sdx]["clean_cdx"].index(cdx)]
+                start = round(char_seg.start * ratio + t1, 3)
+                end = round(char_seg.end * ratio + t1, 3)
+                score = round(char_seg.score, 3)
+
+            char_segments_arr.append(
+                {
+                    "char": char,
+                    "start": start,
+                    "end": end,
+                    "score": score,
+                    "word-idx": word_idx,
+                }
+            )
+
+            # increment word_idx, nltk word tokenization would probably be more robust here, but us space for now...
+            if model_lang in LANGUAGES_WITHOUT_SPACES:
+                word_idx += 1
+            elif cdx == len(text) - 1 or text[cdx+1] == " ":
+                word_idx += 1
+            
+        char_segments_arr = pd.DataFrame(char_segments_arr)
+
+        aligned_subsegments = []
+        # assign sentence_idx to each character index
+        char_segments_arr["sentence-idx"] = None
+        for sdx2, (sstart, send) in enumerate(segment_data[sdx]["sentence_spans"]):
+            curr_chars = char_segments_arr.loc[(char_segments_arr.index >= sstart) & (char_segments_arr.index <= send)]
+            char_segments_arr.loc[(char_segments_arr.index >= sstart) & (char_segments_arr.index <= send), "sentence-idx"] = sdx2
+
+            sentence_text = text[sstart:send]
+            sentence_start = curr_chars["start"].min()
+            end_chars = curr_chars[curr_chars["char"] != ' ']
+            sentence_end = end_chars["end"].max()
+            sentence_words = []
+
+            for word_idx in curr_chars["word-idx"].unique():
+                word_chars = curr_chars.loc[curr_chars["word-idx"] == word_idx]
+                word_text = "".join(word_chars["char"].tolist()).strip()
+                if len(word_text) == 0:
+                    continue
+
+                # dont use space character for alignment
+                word_chars = word_chars[word_chars["char"] != " "]
+
+                word_start = word_chars["start"].min()
+                word_end = word_chars["end"].max()
+                word_score = round(word_chars["score"].mean(), 3)
+
+                # -1 indicates unalignable 
+                word_segment = {"word": word_text}
+
+                if not np.isnan(word_start):
+                    word_segment["start"] = word_start
+                if not np.isnan(word_end):
+                    word_segment["end"] = word_end
+                if not np.isnan(word_score):
+                    word_segment["score"] = word_score
+
+                sentence_words.append(word_segment)
+            
+            aligned_subsegments.append({
+                "text": sentence_text,
+                "start": sentence_start,
+                "end": sentence_end,
+                "words": sentence_words,
+            })
+
+            if return_char_alignments:
+                curr_chars = curr_chars[["char", "start", "end", "score"]]
+                curr_chars.fillna(-1, inplace=True)
+                curr_chars = curr_chars.to_dict("records")
+                curr_chars = [{key: val for key, val in char.items() if val != -1} for char in curr_chars]
+                aligned_subsegments[-1]["chars"] = curr_chars
+
+        aligned_subsegments = pd.DataFrame(aligned_subsegments)
+        aligned_subsegments["start"] = interpolate_nans(aligned_subsegments["start"], method=interpolate_method)
+        aligned_subsegments["end"] = interpolate_nans(aligned_subsegments["end"], method=interpolate_method)
+        # concatenate sentences with same timestamps
+        agg_dict = {"text": " ".join, "words": "sum"}
+        if model_lang in LANGUAGES_WITHOUT_SPACES:
+            agg_dict["text"] = "".join
+        if return_char_alignments:
+            agg_dict["chars"] = "sum"
+        aligned_subsegments= aligned_subsegments.groupby(["start", "end"], as_index=False).agg(agg_dict)
+        aligned_subsegments = aligned_subsegments.to_dict('records')
+        aligned_segments += aligned_subsegments
+
+    # create word_segments list
+    word_segments: List[SingleWordSegment] = []
+    for segment in aligned_segments:
+        word_segments += segment["words"]
+
+    return {"segments": aligned_segments, "word_segments": word_segments}
 
 """
 source: https://pytorch.org/tutorials/intermediate/forced_alignment_with_torchaudio_tutorial.html
 """
+
+
 def get_trellis(emission, tokens, blank_id=0):
     num_frame = emission.size(0)
     num_tokens = len(tokens)
 
-    # Trellis has extra diemsions for both time axis and tokens.
-    # The extra dim for tokens represents <SoS> (start-of-sentence)
-    # The extra dim for time axis is for simplification of the code.
-    trellis = torch.empty((num_frame + 1, num_tokens + 1))
-    trellis[0, 0] = 0
-    trellis[1:, 0] = torch.cumsum(emission[:, 0], 0)
-    trellis[0, -num_tokens:] = -float("inf")
-    trellis[-num_tokens:, 0] = float("inf")
+    trellis = torch.zeros((num_frame, num_tokens))
+    trellis[1:, 0] = torch.cumsum(emission[1:, blank_id], 0)
+    trellis[0, 1:] = -float("inf")
+    trellis[-num_tokens + 1:, 0] = float("inf")
 
-    for t in range(num_frame):
+    for t in range(num_frame - 1):
         trellis[t + 1, 1:] = torch.maximum(
             # Score for staying at the same token
             trellis[t, 1:] + emission[t, blank_id],
             # Score for changing to the next token
-            trellis[t, :-1] + emission[t, tokens],
+            # trellis[t, :-1] + emission[t, tokens[1:]],
+            trellis[t, :-1] + get_wildcard_emission(emission[t], tokens[1:], blank_id),
         )
     return trellis
 
+
+def get_wildcard_emission(frame_emission, tokens, blank_id):
+    """Processing token emission scores containing wildcards (vectorized version)
+
+    Args:
+        frame_emission: Emission probability vector for the current frame
+        tokens: List of token indices
+        blank_id: ID of the blank token
+
+    Returns:
+        tensor: Maximum probability score for each token position
+    """
+    assert 0 <= blank_id < len(frame_emission)
+
+    # Convert tokens to a tensor if they are not already
+    tokens = torch.tensor(tokens) if not isinstance(tokens, torch.Tensor) else tokens
+
+    # Create a mask to identify wildcard positions
+    wildcard_mask = (tokens == -1)
+
+    # Get scores for non-wildcard positions
+    regular_scores = frame_emission[tokens.clamp(min=0)]  # clamp to avoid -1 index
+
+    # Create a mask and compute the maximum value without modifying frame_emission
+    max_valid_score = frame_emission.clone()   # Create a copy
+    max_valid_score[blank_id] = float('-inf')  # Modify the copy to exclude the blank token
+    max_valid_score = max_valid_score.max()
+
+    # Use where operation to combine results
+    result = torch.where(wildcard_mask, max_valid_score, regular_scores)
+
+    return result
+
+
 @dataclass
 class Point:
     token_index: int
     time_index: int
     score: float
 
+
 def backtrack(trellis, emission, tokens, blank_id=0):
-    # Note:
-    # j and t are indices for trellis, which has extra dimensions
-    # for time and tokens at the beginning.
-    # When referring to time frame index `T` in trellis,
-    # the corresponding index in emission is `T-1`.
-    # Similarly, when referring to token index `J` in trellis,
-    # the corresponding index in transcript is `J-1`.
-    j = trellis.size(1) - 1
-    t_start = torch.argmax(trellis[:, j]).item()
+    t, j = trellis.size(0) - 1, trellis.size(1) - 1
+
+    path = [Point(j, t, emission[t, blank_id].exp().item())]
+    while j > 0:
+        # Should not happen but just in case
+        assert t > 0
 
-    path = []
-    for t in range(t_start, 0, -1):
         # 1. Figure out if the current position was stay or change
-        # Note (again):
-        # `emission[J-1]` is the emission at time frame `J` of trellis dimension.
-        # Score for token staying the same from time frame J-1 to T.
-        stayed = trellis[t - 1, j] + emission[t - 1, blank_id]
-        # Score for token changing from C-1 at T-1 to J at T.
-        changed = trellis[t - 1, j - 1] + emission[t - 1, tokens[j - 1]]
+        # Frame-wise score of stay vs change
+        p_stay = emission[t - 1, blank_id]
+        # p_change = emission[t - 1, tokens[j]]
+        p_change = get_wildcard_emission(emission[t - 1], [tokens[j]], blank_id)[0]
 
-        # 2. Store the path with frame-wise probability.
-        prob = emission[t - 1, tokens[j - 1] if changed > stayed else 0].exp().item()
-        # Return token index and time index in non-trellis coordinate.
-        path.append(Point(j - 1, t - 1, prob))
+        # Context-aware score for stay vs change
+        stayed = trellis[t - 1, j] + p_stay
+        changed = trellis[t - 1, j - 1] + p_change
 
-        # 3. Update the token
+        # Update position
+        t -= 1
         if changed > stayed:
             j -= 1
-            if j == 0:
-                break
-    else:
-        # failed
-        return None
+
+        # Store the path with frame-wise probability.
+        prob = (p_change if changed > stayed else p_stay).exp().item()
+        path.append(Point(j, t, prob))
+
+    # Now j == 0, which means, it reached the SoS.
+    # Fill up the rest for the sake of visualization
+    while t > 0:
+        prob = emission[t - 1, blank_id].exp().item()
+        path.append(Point(j, t - 1, prob))
+        t -= 1
+
     return path[::-1]
 
+
+
+@dataclass
+class Path:
+    points: List[Point]
+    score: float
+
+
+@dataclass
+class BeamState:
+    """State in beam search."""
+    token_index: int   # Current token position
+    time_index: int    # Current time step
+    score: float       # Cumulative score
+    path: List[Point]  # Path history
+
+
+def backtrack_beam(trellis, emission, tokens, blank_id=0, beam_width=5):
+    """Standard CTC beam search backtracking implementation.
+
+    Args:
+        trellis (torch.Tensor): The trellis (or lattice) of shape (T, N), where T is the number of time steps
+                                and N is the number of tokens (including the blank token).
+        emission (torch.Tensor): The emission probabilities of shape (T, N).
+        tokens (List[int]): List of token indices (excluding the blank token).
+        blank_id (int, optional): The ID of the blank token. Defaults to 0.
+        beam_width (int, optional): The number of top paths to keep during beam search. Defaults to 5.
+
+    Returns:
+        List[Point]: the best path
+    """
+    T, J = trellis.size(0) - 1, trellis.size(1) - 1
+
+    init_state = BeamState(
+        token_index=J,
+        time_index=T,
+        score=trellis[T, J],
+        path=[Point(J, T, emission[T, blank_id].exp().item())]
+    )
+
+    beams = [init_state]
+
+    while beams and beams[0].token_index > 0:
+        next_beams = []
+
+        for beam in beams:
+            t, j = beam.time_index, beam.token_index
+
+            if t <= 0:
+                continue
+
+            p_stay = emission[t - 1, blank_id]
+            p_change = get_wildcard_emission(emission[t - 1], [tokens[j]], blank_id)[0]
+
+            stay_score = trellis[t - 1, j]
+            change_score = trellis[t - 1, j - 1] if j > 0 else float('-inf')
+
+            # Stay
+            if not math.isinf(stay_score):
+                new_path = beam.path.copy()
+                new_path.append(Point(j, t - 1, p_stay.exp().item()))
+                next_beams.append(BeamState(
+                    token_index=j,
+                    time_index=t - 1,
+                    score=stay_score,
+                    path=new_path
+                ))
+
+            # Change
+            if j > 0 and not math.isinf(change_score):
+                new_path = beam.path.copy()
+                new_path.append(Point(j - 1, t - 1, p_change.exp().item()))
+                next_beams.append(BeamState(
+                    token_index=j - 1,
+                    time_index=t - 1,
+                    score=change_score,
+                    path=new_path
+                ))
+
+        # sort by score
+        beams = sorted(next_beams, key=lambda x: x.score, reverse=True)[:beam_width]
+
+        if not beams:
+            break
+
+    if not beams:
+        return None
+
+    best_beam = beams[0]
+    t = best_beam.time_index
+    j = best_beam.token_index
+    while t > 0:
+        prob = emission[t - 1, blank_id].exp().item()
+        best_beam.path.append(Point(j, t - 1, prob))
+        t -= 1
+
+    return best_beam.path[::-1]
+
+
 # Merge the labels
 @dataclass
 class Segment:

whisperx/asr.py

@@ -0,0 +1,416 @@
+import os
+from typing import List, Optional, Union
+from dataclasses import replace
+
+import ctranslate2
+import faster_whisper
+import numpy as np
+import torch
+from faster_whisper.tokenizer import Tokenizer
+from faster_whisper.transcribe import TranscriptionOptions, get_ctranslate2_storage
+from transformers import Pipeline
+from transformers.pipelines.pt_utils import PipelineIterator
+
+from whisperx.audio import N_SAMPLES, SAMPLE_RATE, load_audio, log_mel_spectrogram
+from whisperx.types import SingleSegment, TranscriptionResult
+from whisperx.vads import Vad, Silero, Pyannote
+
+
+def find_numeral_symbol_tokens(tokenizer):
+    numeral_symbol_tokens = []
+    for i in range(tokenizer.eot):
+        token = tokenizer.decode([i]).removeprefix(" ")
+        has_numeral_symbol = any(c in "0123456789%$£" for c in token)
+        if has_numeral_symbol:
+            numeral_symbol_tokens.append(i)
+    return numeral_symbol_tokens
+
+class WhisperModel(faster_whisper.WhisperModel):
+    '''
+    FasterWhisperModel provides batched inference for faster-whisper.
+    Currently only works in non-timestamp mode and fixed prompt for all samples in batch.
+    '''
+
+    def generate_segment_batched(
+        self,
+        features: np.ndarray,
+        tokenizer: Tokenizer,
+        options: TranscriptionOptions,
+        encoder_output=None,
+    ):
+        batch_size = features.shape[0]
+        all_tokens = []
+        prompt_reset_since = 0
+        if options.initial_prompt is not None:
+            initial_prompt = " " + options.initial_prompt.strip()
+            initial_prompt_tokens = tokenizer.encode(initial_prompt)
+            all_tokens.extend(initial_prompt_tokens)
+        previous_tokens = all_tokens[prompt_reset_since:]
+        prompt = self.get_prompt(
+            tokenizer,
+            previous_tokens,
+            without_timestamps=options.without_timestamps,
+            prefix=options.prefix,
+            hotwords=options.hotwords
+        )
+
+        encoder_output = self.encode(features)
+
+        max_initial_timestamp_index = int(
+            round(options.max_initial_timestamp / self.time_precision)
+        )
+
+        result = self.model.generate(
+                encoder_output,
+                [prompt] * batch_size,
+                beam_size=options.beam_size,
+                patience=options.patience,
+                length_penalty=options.length_penalty,
+                max_length=self.max_length,
+                suppress_blank=options.suppress_blank,
+                suppress_tokens=options.suppress_tokens,
+            )
+
+        tokens_batch = [x.sequences_ids[0] for x in result]
+
+        def decode_batch(tokens: List[List[int]]) -> str:
+            res = []
+            for tk in tokens:
+                res.append([token for token in tk if token < tokenizer.eot])
+            # text_tokens = [token for token in tokens if token < self.eot]
+            return tokenizer.tokenizer.decode_batch(res)
+
+        text = decode_batch(tokens_batch)
+
+        return text
+
+    def encode(self, features: np.ndarray) -> ctranslate2.StorageView:
+        # When the model is running on multiple GPUs, the encoder output should be moved
+        # to the CPU since we don't know which GPU will handle the next job.
+        to_cpu = self.model.device == "cuda" and len(self.model.device_index) > 1
+        # unsqueeze if batch size = 1
+        if len(features.shape) == 2:
+            features = np.expand_dims(features, 0)
+        features = get_ctranslate2_storage(features)
+
+        return self.model.encode(features, to_cpu=to_cpu)
+
+class FasterWhisperPipeline(Pipeline):
+    """
+    Huggingface Pipeline wrapper for FasterWhisperModel.
+    """
+    # TODO:
+    # - add support for timestamp mode
+    # - add support for custom inference kwargs
+
+    def __init__(
+        self,
+        model: WhisperModel,
+        vad,
+        vad_params: dict,
+        options: TranscriptionOptions,
+        tokenizer: Optional[Tokenizer] = None,
+        device: Union[int, str, "torch.device"] = -1,
+        framework="pt",
+        language: Optional[str] = None,
+        suppress_numerals: bool = False,
+        **kwargs,
+    ):
+        self.model = model
+        self.tokenizer = tokenizer
+        self.options = options
+        self.preset_language = language
+        self.suppress_numerals = suppress_numerals
+        self._batch_size = kwargs.pop("batch_size", None)
+        self._num_workers = 1
+        self._preprocess_params, self._forward_params, self._postprocess_params = self._sanitize_parameters(**kwargs)
+        self.call_count = 0
+        self.framework = framework
+        if self.framework == "pt":
+            if isinstance(device, torch.device):
+                self.device = device
+            elif isinstance(device, str):
+                self.device = torch.device(device)
+            elif device < 0:
+                self.device = torch.device("cpu")
+            else:
+                self.device = torch.device(f"cuda:{device}")
+        else:
+            self.device = device
+
+        super(Pipeline, self).__init__()
+        self.vad_model = vad
+        self._vad_params = vad_params
+
+    def _sanitize_parameters(self, **kwargs):
+        preprocess_kwargs = {}
+        if "tokenizer" in kwargs:
+            preprocess_kwargs["maybe_arg"] = kwargs["maybe_arg"]
+        return preprocess_kwargs, {}, {}
+
+    def preprocess(self, audio):
+        audio = audio['inputs']
+        model_n_mels = self.model.feat_kwargs.get("feature_size")
+        features = log_mel_spectrogram(
+            audio,
+            n_mels=model_n_mels if model_n_mels is not None else 80,
+            padding=N_SAMPLES - audio.shape[0],
+        )
+        return {'inputs': features}
+
+    def _forward(self, model_inputs):
+        outputs = self.model.generate_segment_batched(model_inputs['inputs'], self.tokenizer, self.options)
+        return {'text': outputs}
+
+    def postprocess(self, model_outputs):
+        return model_outputs
+
+    def get_iterator(
+        self,
+        inputs,
+        num_workers: int,
+        batch_size: int,
+        preprocess_params: dict,
+        forward_params: dict,
+        postprocess_params: dict,
+    ):
+        dataset = PipelineIterator(inputs, self.preprocess, preprocess_params)
+        if "TOKENIZERS_PARALLELISM" not in os.environ:
+            os.environ["TOKENIZERS_PARALLELISM"] = "false"
+        # TODO hack by collating feature_extractor and image_processor
+
+        def stack(items):
+            return {'inputs': torch.stack([x['inputs'] for x in items])}
+        dataloader = torch.utils.data.DataLoader(dataset, num_workers=num_workers, batch_size=batch_size, collate_fn=stack)
+        model_iterator = PipelineIterator(dataloader, self.forward, forward_params, loader_batch_size=batch_size)
+        final_iterator = PipelineIterator(model_iterator, self.postprocess, postprocess_params)
+        return final_iterator
+
+    def transcribe(
+        self,
+        audio: Union[str, np.ndarray],
+        batch_size: Optional[int] = None,
+        num_workers=0,
+        language: Optional[str] = None,
+        task: Optional[str] = None,
+        chunk_size=30,
+        print_progress=False,
+        combined_progress=False,
+        verbose=False,
+    ) -> TranscriptionResult:
+        if isinstance(audio, str):
+            audio = load_audio(audio)
+
+        def data(audio, segments):
+            for seg in segments:
+                f1 = int(seg['start'] * SAMPLE_RATE)
+                f2 = int(seg['end'] * SAMPLE_RATE)
+                # print(f2-f1)
+                yield {'inputs': audio[f1:f2]}
+
+        # Pre-process audio and merge chunks as defined by the respective VAD child class 
+        # In case vad_model is manually assigned (see 'load_model') follow the functionality of pyannote toolkit
+        if issubclass(type(self.vad_model), Vad):
+            waveform = self.vad_model.preprocess_audio(audio)
+            merge_chunks =  self.vad_model.merge_chunks
+        else:
+            waveform = Pyannote.preprocess_audio(audio)
+            merge_chunks = Pyannote.merge_chunks
+
+        vad_segments = self.vad_model({"waveform": waveform, "sample_rate": SAMPLE_RATE})
+        vad_segments = merge_chunks(
+            vad_segments,
+            chunk_size,
+            onset=self._vad_params["vad_onset"],
+            offset=self._vad_params["vad_offset"],
+        )
+        if self.tokenizer is None:
+            language = language or self.detect_language(audio)
+            task = task or "transcribe"
+            self.tokenizer = Tokenizer(
+                self.model.hf_tokenizer,
+                self.model.model.is_multilingual,
+                task=task,
+                language=language,
+            )
+        else:
+            language = language or self.tokenizer.language_code
+            task = task or self.tokenizer.task
+            if task != self.tokenizer.task or language != self.tokenizer.language_code:
+                self.tokenizer = Tokenizer(
+                    self.model.hf_tokenizer,
+                    self.model.model.is_multilingual,
+                    task=task,
+                    language=language,
+                )
+
+        if self.suppress_numerals:
+            previous_suppress_tokens = self.options.suppress_tokens
+            numeral_symbol_tokens = find_numeral_symbol_tokens(self.tokenizer)
+            print(f"Suppressing numeral and symbol tokens")
+            new_suppressed_tokens = numeral_symbol_tokens + self.options.suppress_tokens
+            new_suppressed_tokens = list(set(new_suppressed_tokens))
+            self.options = replace(self.options, suppress_tokens=new_suppressed_tokens)
+
+        segments: List[SingleSegment] = []
+        batch_size = batch_size or self._batch_size
+        total_segments = len(vad_segments)
+        for idx, out in enumerate(self.__call__(data(audio, vad_segments), batch_size=batch_size, num_workers=num_workers)):
+            if print_progress:
+                base_progress = ((idx + 1) / total_segments) * 100
+                percent_complete = base_progress / 2 if combined_progress else base_progress
+                print(f"Progress: {percent_complete:.2f}%...")
+            text = out['text']
+            if batch_size in [0, 1, None]:
+                text = text[0]
+            if verbose:
+                print(f"Transcript: [{round(vad_segments[idx]['start'], 3)} --> {round(vad_segments[idx]['end'], 3)}] {text}")
+            segments.append(
+                {
+                    "text": text,
+                    "start": round(vad_segments[idx]['start'], 3),
+                    "end": round(vad_segments[idx]['end'], 3)
+                }
+            )
+
+        # revert the tokenizer if multilingual inference is enabled
+        if self.preset_language is None:
+            self.tokenizer = None
+
+        # revert suppressed tokens if suppress_numerals is enabled
+        if self.suppress_numerals:
+            self.options = replace(self.options, suppress_tokens=previous_suppress_tokens)
+
+        return {"segments": segments, "language": language}
+
+    def detect_language(self, audio: np.ndarray) -> str:
+        if audio.shape[0] < N_SAMPLES:
+            print("Warning: audio is shorter than 30s, language detection may be inaccurate.")
+        model_n_mels = self.model.feat_kwargs.get("feature_size")
+        segment = log_mel_spectrogram(audio[: N_SAMPLES],
+                                      n_mels=model_n_mels if model_n_mels is not None else 80,
+                                      padding=0 if audio.shape[0] >= N_SAMPLES else N_SAMPLES - audio.shape[0])
+        encoder_output = self.model.encode(segment)
+        results = self.model.model.detect_language(encoder_output)
+        language_token, language_probability = results[0][0]
+        language = language_token[2:-2]
+        print(f"Detected language: {language} ({language_probability:.2f}) in first 30s of audio...")
+        return language
+
+
+def load_model(
+    whisper_arch: str,
+    device: str,
+    device_index=0,
+    compute_type="float16",
+    asr_options: Optional[dict] = None,
+    language: Optional[str] = None,
+    vad_model: Optional[Vad]= None,
+    vad_method: Optional[str] = "pyannote",
+    vad_options: Optional[dict] = None,
+    model: Optional[WhisperModel] = None,
+    task="transcribe",
+    download_root: Optional[str] = None,
+    local_files_only=False,
+    threads=4,
+) -> FasterWhisperPipeline:
+    """Load a Whisper model for inference.
+    Args:
+        whisper_arch - The name of the Whisper model to load.
+        device - The device to load the model on.
+        compute_type - The compute type to use for the model.
+        vad_method - The vad method to use. vad_model has higher priority if is not None.
+        options - A dictionary of options to use for the model.
+        language - The language of the model. (use English for now)
+        model - The WhisperModel instance to use.
+        download_root - The root directory to download the model to.
+        local_files_only - If `True`, avoid downloading the file and return the path to the local cached file if it exists.
+        threads - The number of cpu threads to use per worker, e.g. will be multiplied by num workers.
+    Returns:
+        A Whisper pipeline.
+    """
+
+    if whisper_arch.endswith(".en"):
+        language = "en"
+
+    model = model or WhisperModel(whisper_arch,
+                         device=device,
+                         device_index=device_index,
+                         compute_type=compute_type,
+                         download_root=download_root,
+                         local_files_only=local_files_only,
+                         cpu_threads=threads)
+    if language is not None:
+        tokenizer = Tokenizer(model.hf_tokenizer, model.model.is_multilingual, task=task, language=language)
+    else:
+        print("No language specified, language will be first be detected for each audio file (increases inference time).")
+        tokenizer = None
+
+    default_asr_options =  {
+        "beam_size": 5,
+        "best_of": 5,
+        "patience": 1,
+        "length_penalty": 1,
+        "repetition_penalty": 1,
+        "no_repeat_ngram_size": 0,
+        "temperatures": [0.0, 0.2, 0.4, 0.6, 0.8, 1.0],
+        "compression_ratio_threshold": 2.4,
+        "log_prob_threshold": -1.0,
+        "no_speech_threshold": 0.6,
+        "condition_on_previous_text": False,
+        "prompt_reset_on_temperature": 0.5,
+        "initial_prompt": None,
+        "prefix": None,
+        "suppress_blank": True,
+        "suppress_tokens": [-1],
+        "without_timestamps": True,
+        "max_initial_timestamp": 0.0,
+        "word_timestamps": False,
+        "prepend_punctuations": "\"'“¿([{-",
+        "append_punctuations": "\"'.。,，!！?？:：”)]}、",
+        "multilingual": model.model.is_multilingual,
+        "suppress_numerals": False,
+        "max_new_tokens": None,
+        "clip_timestamps": None,
+        "hallucination_silence_threshold": None,
+        "hotwords": None,
+    }
+
+    if asr_options is not None:
+        default_asr_options.update(asr_options)
+
+    suppress_numerals = default_asr_options["suppress_numerals"]
+    del default_asr_options["suppress_numerals"]
+
+    default_asr_options = TranscriptionOptions(**default_asr_options)
+
+    default_vad_options = {
+        "chunk_size": 30, # needed by silero since binarization happens before merge_chunks
+        "vad_onset": 0.500,
+        "vad_offset": 0.363
+    }
+
+    if vad_options is not None:
+        default_vad_options.update(vad_options)
+
+    # Note: manually assigned vad_model has higher priority than vad_method!
+    if vad_model is not None:
+        print("Use manually assigned vad_model. vad_method is ignored.")
+        vad_model = vad_model
+    else:
+        if vad_method == "silero":
+            vad_model = Silero(**default_vad_options)
+        elif vad_method == "pyannote":
+            vad_model = Pyannote(torch.device(device), use_auth_token=None, **default_vad_options)
+        else:
+            raise ValueError(f"Invalid vad_method: {vad_method}")
+
+    return FasterWhisperPipeline(
+        model=model,
+        vad=vad_model,
+        options=default_asr_options,
+        tokenizer=tokenizer,
+        language=language,
+        suppress_numerals=suppress_numerals,
+        vad_params=default_vad_options,
+    )

whisperx/assets/gpt2/special_tokens_map.json

@@ -1 +0,0 @@
-{"bos_token": "<|endoftext|>", "eos_token": "<|endoftext|>", "unk_token": "<|endoftext|>"}

whisperx/assets/gpt2/tokenizer_config.json

@@ -1 +0,0 @@
-{"unk_token": "<|endoftext|>", "bos_token": "<|endoftext|>", "eos_token": "<|endoftext|>", "add_prefix_space": false, "model_max_length": 1024, "special_tokens_map_file": null, "name_or_path": "gpt2", "tokenizer_class": "GPT2Tokenizer"}

whisperx/assets/multilingual/added_tokens.json

@@ -1 +0,0 @@
-{"<|endoftext|>": 50257}

whisperx/assets/multilingual/special_tokens_map.json

@@ -1 +0,0 @@
-{"bos_token": "<|endoftext|>", "eos_token": "<|endoftext|>", "unk_token": "<|endoftext|>"}

whisperx/assets/multilingual/tokenizer_config.json

@@ -1 +0,0 @@
-{"unk_token": {"content": "<|endoftext|>", "single_word": false, "lstrip": false, "rstrip": false, "normalized": true, "__type": "AddedToken"}, "bos_token": {"content": "<|endoftext|>", "single_word": false, "lstrip": false, "rstrip": false, "normalized": true, "__type": "AddedToken"}, "eos_token": {"content": "<|endoftext|>", "single_word": false, "lstrip": false, "rstrip": false, "normalized": true, "__type": "AddedToken"}, "add_prefix_space": false, "model_max_length": 1024, "special_tokens_map_file": null, "name_or_path": "multilingual", "errors": "replace", "tokenizer_class": "GPT2Tokenizer"}

whisperx/audio.py

@@ -1,25 +1,28 @@
 import os
+import subprocess
 from functools import lru_cache
-from typing import Union
+from typing import Optional, Union
 
-import ffmpeg
 import numpy as np
 import torch
 import torch.nn.functional as F
 
-from .utils import exact_div
+from whisperx.utils import exact_div
 
 # hard-coded audio hyperparameters
 SAMPLE_RATE = 16000
 N_FFT = 400
-N_MELS = 80
 HOP_LENGTH = 160
 CHUNK_LENGTH = 30
-N_SAMPLES = CHUNK_LENGTH * SAMPLE_RATE  # 480000: number of samples in a chunk
-N_FRAMES = exact_div(N_SAMPLES, HOP_LENGTH)  # 3000: number of frames in a mel spectrogram input
+N_SAMPLES = CHUNK_LENGTH * SAMPLE_RATE  # 480000 samples in a 30-second chunk
+N_FRAMES = exact_div(N_SAMPLES, HOP_LENGTH)  # 3000 frames in a mel spectrogram input
+
+N_SAMPLES_PER_TOKEN = HOP_LENGTH * 2  # the initial convolutions has stride 2
+FRAMES_PER_SECOND = exact_div(SAMPLE_RATE, HOP_LENGTH)  # 10ms per audio frame
+TOKENS_PER_SECOND = exact_div(SAMPLE_RATE, N_SAMPLES_PER_TOKEN)  # 20ms per audio token
 
 
-def load_audio(file: str, sr: int = SAMPLE_RATE):
+def load_audio(file: str, sr: int = SAMPLE_RATE) -> np.ndarray:
     """
     Open an audio file and read as mono waveform, resampling as necessary
 
@@ -36,14 +39,27 @@ def load_audio(file: str, sr: int = SAMPLE_RATE):
     A NumPy array containing the audio waveform, in float32 dtype.
     """
     try:
-        # This launches a subprocess to decode audio while down-mixing and resampling as necessary.
-        # Requires the ffmpeg CLI and `ffmpeg-python` package to be installed.
-        out, _ = (
-            ffmpeg.input(file, threads=0)
-            .output("-", format="s16le", acodec="pcm_s16le", ac=1, ar=sr)
-            .run(cmd=["ffmpeg", "-nostdin"], capture_stdout=True, capture_stderr=True)
-        )
-    except ffmpeg.Error as e:
+        # Launches a subprocess to decode audio while down-mixing and resampling as necessary.
+        # Requires the ffmpeg CLI to be installed.
+        cmd = [
+            "ffmpeg",
+            "-nostdin",
+            "-threads",
+            "0",
+            "-i",
+            file,
+            "-f",
+            "s16le",
+            "-ac",
+            "1",
+            "-acodec",
+            "pcm_s16le",
+            "-ar",
+            str(sr),
+            "-",
+        ]
+        out = subprocess.run(cmd, capture_output=True, check=True).stdout
+    except subprocess.CalledProcessError as e:
         raise RuntimeError(f"Failed to load audio: {e.stderr.decode()}") from e
 
     return np.frombuffer(out, np.int16).flatten().astype(np.float32) / 32768.0
@@ -55,7 +71,9 @@ def pad_or_trim(array, length: int = N_SAMPLES, *, axis: int = -1):
     """
     if torch.is_tensor(array):
         if array.shape[axis] > length:
-            array = array.index_select(dim=axis, index=torch.arange(length, device=array.device))
+            array = array.index_select(
+                dim=axis, index=torch.arange(length, device=array.device)
+            )
 
         if array.shape[axis] < length:
             pad_widths = [(0, 0)] * array.ndim
@@ -74,7 +92,7 @@ def pad_or_trim(array, length: int = N_SAMPLES, *, axis: int = -1):
 
 
 @lru_cache(maxsize=None)
-def mel_filters(device, n_mels: int = N_MELS) -> torch.Tensor:
+def mel_filters(device, n_mels: int) -> torch.Tensor:
     """
     load the mel filterbank matrix for projecting STFT into a Mel spectrogram.
     Allows decoupling librosa dependency; saved using:
@@ -84,12 +102,19 @@ def mel_filters(device, n_mels: int = N_MELS) -> torch.Tensor:
             mel_80=librosa.filters.mel(sr=16000, n_fft=400, n_mels=80),
         )
     """
-    assert n_mels == 80, f"Unsupported n_mels: {n_mels}"
-    with np.load(os.path.join(os.path.dirname(__file__), "assets", "mel_filters.npz")) as f:
+    assert n_mels in [80, 128], f"Unsupported n_mels: {n_mels}"
+    with np.load(
+        os.path.join(os.path.dirname(__file__), "assets", "mel_filters.npz")
+    ) as f:
         return torch.from_numpy(f[f"mel_{n_mels}"]).to(device)
 
 
-def log_mel_spectrogram(audio: Union[str, np.ndarray, torch.Tensor], n_mels: int = N_MELS):
+def log_mel_spectrogram(
+    audio: Union[str, np.ndarray, torch.Tensor],
+    n_mels: int,
+    padding: int = 0,
+    device: Optional[Union[str, torch.device]] = None,
+):
     """
     Compute the log-Mel spectrogram of
 
@@ -101,6 +126,12 @@ def log_mel_spectrogram(audio: Union[str, np.ndarray, torch.Tensor], n_mels: int
     n_mels: int
         The number of Mel-frequency filters, only 80 is supported
 
+    padding: int
+        Number of zero samples to pad to the right
+
+    device: Optional[Union[str, torch.device]]
+        If given, the audio tensor is moved to this device before STFT
+
     Returns
     -------
     torch.Tensor, shape = (80, n_frames)
@@ -111,6 +142,10 @@ def log_mel_spectrogram(audio: Union[str, np.ndarray, torch.Tensor], n_mels: int
             audio = load_audio(audio)
         audio = torch.from_numpy(audio)
 
+    if device is not None:
+        audio = audio.to(device)
+    if padding > 0:
+        audio = F.pad(audio, (0, padding))
     window = torch.hann_window(N_FFT).to(audio.device)
     stft = torch.stft(audio, N_FFT, HOP_LENGTH, window=window, return_complex=True)
     magnitudes = stft[..., :-1].abs() ** 2
@@ -121,4 +156,4 @@ def log_mel_spectrogram(audio: Union[str, np.ndarray, torch.Tensor], n_mels: int
     log_spec = torch.clamp(mel_spec, min=1e-10).log10()
     log_spec = torch.maximum(log_spec, log_spec.max() - 8.0)
     log_spec = (log_spec + 4.0) / 4.0
-    return log_spec
+    return log_spec

whisperx/conjunctions.py

@@ -0,0 +1,47 @@
+# conjunctions.py
+
+from typing import Set
+
+
+conjunctions_by_language = {
+    'en': {'and', 'whether', 'or', 'as', 'but', 'so', 'for', 'nor', 'which', 'yet', 'although', 'since', 'unless', 'when', 'while', 'because', 'if', 'how', 'that', 'than', 'who', 'where', 'what', 'near', 'before', 'after', 'across', 'through', 'until', 'once', 'whereas', 'even', 'both', 'either', 'neither', 'though'},
+    'fr': {'et', 'ou', 'mais', 'parce', 'bien', 'pendant', 'quand', 'où', 'comme', 'si', 'que', 'avant', 'après', 'aussitôt', 'jusqu’à', 'à', 'malgré', 'donc', 'tant', 'puisque', 'ni', 'soit', 'bien', 'encore', 'dès', 'lorsque'},
+    'de': {'und', 'oder', 'aber', 'weil', 'obwohl', 'während', 'wenn', 'wo', 'wie', 'dass', 'bevor', 'nachdem', 'sobald', 'bis', 'außer', 'trotzdem', 'also', 'sowie', 'indem', 'weder', 'sowohl', 'zwar', 'jedoch'},
+    'es': {'y', 'o', 'pero', 'porque', 'aunque', 'sin', 'mientras', 'cuando', 'donde', 'como', 'si', 'que', 'antes', 'después', 'tan', 'hasta', 'a', 'a', 'por', 'ya', 'ni', 'sino'},
+    'it': {'e', 'o', 'ma', 'perché', 'anche', 'mentre', 'quando', 'dove', 'come', 'se', 'che', 'prima', 'dopo', 'appena', 'fino', 'a', 'nonostante', 'quindi', 'poiché', 'né', 'ossia', 'cioè'},
+    'ja': {'そして', 'または', 'しかし', 'なぜなら', 'もし', 'それとも', 'だから', 'それに', 'なのに', 'そのため', 'かつ', 'それゆえに', 'ならば', 'もしくは', 'ため'},
+    'zh': {'和', '或', '但是', '因为', '任何', '也', '虽然', '而且', '所以', '如果', '除非', '尽管', '既然', '即使', '只要', '直到', '然后', '因此', '不但', '而是', '不过'},
+    'nl': {'en', 'of', 'maar', 'omdat', 'hoewel', 'terwijl', 'wanneer', 'waar', 'zoals', 'als', 'dat', 'voordat', 'nadat', 'zodra', 'totdat', 'tenzij', 'ondanks', 'dus', 'zowel', 'noch', 'echter', 'toch'},
+    'uk': {'та', 'або', 'але', 'тому', 'хоча', 'поки', 'бо', 'коли', 'де', 'як', 'якщо', 'що', 'перш', 'після', 'доки', 'незважаючи', 'тому', 'ані'},
+    'pt': {'e', 'ou', 'mas', 'porque', 'embora', 'enquanto', 'quando', 'onde', 'como', 'se', 'que', 'antes', 'depois', 'assim', 'até', 'a', 'apesar', 'portanto', 'já', 'pois', 'nem', 'senão'},
+    'ar': {'و', 'أو', 'لكن', 'لأن', 'مع', 'بينما', 'عندما', 'حيث', 'كما', 'إذا', 'الذي', 'قبل', 'بعد', 'فور', 'حتى', 'إلا', 'رغم', 'لذلك', 'بما'},
+    'cs': {'a', 'nebo', 'ale', 'protože', 'ačkoli', 'zatímco', 'když', 'kde', 'jako', 'pokud', 'že', 'než', 'poté', 'jakmile', 'dokud', 'pokud ne', 'navzdory', 'tak', 'stejně', 'ani', 'tudíž'},
+    'ru': {'и', 'или', 'но', 'потому', 'хотя', 'пока', 'когда', 'где', 'как', 'если', 'что', 'перед', 'после', 'несмотря', 'таким', 'также', 'ни', 'зато'},
+    'pl': {'i', 'lub', 'ale', 'ponieważ', 'chociaż', 'podczas', 'kiedy', 'gdzie', 'jak', 'jeśli', 'że', 'zanim', 'po', 'jak tylko', 'dopóki', 'chyba', 'pomimo', 'więc', 'tak', 'ani', 'czyli'},
+    'hu': {'és', 'vagy', 'de', 'mert', 'habár', 'míg', 'amikor', 'ahol', 'ahogy', 'ha', 'hogy', 'mielőtt', 'miután', 'amint', 'amíg', 'hacsak', 'ellenére', 'tehát', 'úgy', 'sem', 'vagyis'},
+    'fi': {'ja', 'tai', 'mutta', 'koska', 'vaikka', 'kun', 'missä', 'kuten', 'jos', 'että', 'ennen', 'sen jälkeen', 'heti', 'kunnes', 'ellei', 'huolimatta', 'siis', 'sekä', 'eikä', 'vaan'},
+    'fa': {'و', 'یا', 'اما', 'چون', 'اگرچه', 'در حالی', 'وقتی', 'کجا', 'چگونه', 'اگر', 'که', 'قبل', 'پس', 'به محض', 'تا زمانی', 'مگر', 'با وجود', 'پس', 'همچنین', 'نه'},
+    'el': {'και', 'ή', 'αλλά', 'επειδή', 'αν', 'ενώ', 'όταν', 'όπου', 'όπως', 'αν', 'που', 'προτού', 'αφού', 'μόλις', 'μέχρι', 'εκτός', 'παρά', 'έτσι', 'όπως', 'ούτε', 'δηλαδή'},
+    'tr': {'ve', 'veya', 'ama', 'çünkü', 'her ne', 'iken', 'nerede', 'nasıl', 'eğer', 'ki', 'önce', 'sonra', 'hemen', 'kadar', 'rağmen', 'hem', 'ne', 'yani'},
+    'da': {'og', 'eller', 'men', 'fordi', 'selvom', 'mens', 'når', 'hvor', 'som', 'hvis', 'at', 'før', 'efter', 'indtil', 'medmindre', 'således', 'ligesom', 'hverken', 'altså'},
+    'he': {'ו', 'או', 'אבל', 'כי', 'אף', 'בזמן', 'כאשר', 'היכן', 'כיצד', 'אם', 'ש', 'לפני', 'אחרי', 'ברגע', 'עד', 'אלא', 'למרות', 'לכן', 'כמו', 'לא', 'אז'},
+    'vi': {'và', 'hoặc', 'nhưng', 'bởi', 'mặc', 'trong', 'khi', 'ở', 'như', 'nếu', 'rằng', 'trước', 'sau', 'ngay', 'cho', 'trừ', 'mặc', 'vì', 'giống', 'cũng', 'tức'},
+    'ko': {'그리고', '또는','그런데','그래도', '이나', '결국', '마지막으로', '마찬가지로', '반면에', '아니면', '거나', '또는', '그럼에도', '그렇기', '때문에', '덧붙이자면', '게다가', '그러나',  '고', '그래서', '랑', '한다면', '하지만', '무엇', '왜냐하면', '비록', '동안', '언제', '어디서', '어떻게', '만약', '그', '전에', '후에', '즉시', '까지', '아니라면', '불구하고', '따라서', '같은', '도'},
+    'ur': {'اور', 'یا', 'مگر', 'کیونکہ', 'اگرچہ', 'جبکہ', 'جب', 'کہاں', 'کس طرح', 'اگر', 'کہ', 'سے پہلے', 'کے بعد', 'جیسے ہی', 'تک', 'اگر نہیں تو', 'کے باوجود', 'اس لئے', 'جیسے', 'نہ'},
+    'hi': {'और', 'या', 'पर', 'तो', 'न', 'फिर', 'हालांकि', 'चूंकि', 'अगर', 'कैसे', 'वह', 'से', 'जो', 'जहां', 'क्या', 'नजदीक', 'पहले', 'बाद', 'के', 'पार', 'माध्यम', 'तक', 'एक', 'जबकि', 'यहां', 'तक', 'दोनों', 'या', 'न', 'हालांकि'}
+
+}
+
+commas_by_language = {
+    'ja': '、', 
+    'zh': '，',
+    'fa': '،', 
+    'ur': '،'  
+}
+
+def get_conjunctions(lang_code: str) -> Set[str]:
+    return conjunctions_by_language.get(lang_code, set())
+
+
+def get_comma(lang_code: str) -> str:
+    return commas_by_language.get(lang_code, ",")

whisperx/decoding.py

@@ -1,710 +0,0 @@
-from dataclasses import dataclass, field
-from typing import Dict, List, Tuple, Iterable, Optional, Sequence, Union, TYPE_CHECKING
-
-import numpy as np
-import torch
-import torch.nn.functional as F
-from torch import Tensor
-from torch.distributions import Categorical
-
-from .audio import CHUNK_LENGTH
-from .tokenizer import Tokenizer, get_tokenizer
-from .utils import compression_ratio
-
-if TYPE_CHECKING:
-    from .model import Whisper
-
-
-@torch.no_grad()
-def detect_language(model: "Whisper", mel: Tensor, tokenizer: Tokenizer = None) -> Tuple[Tensor, List[dict]]:
-    """
-    Detect the spoken language in the audio, and return them as list of strings, along with the ids
-    of the most probable language tokens and the probability distribution over all language tokens.
-    This is performed outside the main decode loop in order to not interfere with kv-caching.
-
-    Returns
-    -------
-    language_tokens : Tensor, shape = (n_audio,)
-        ids of the most probable language tokens, which appears after the startoftranscript token.
-    language_probs : List[Dict[str, float]], length = n_audio
-        list of dictionaries containing the probability distribution over all languages.
-    """
-    if tokenizer is None:
-        tokenizer = get_tokenizer(model.is_multilingual)
-    if tokenizer.language is None or tokenizer.language_token not in tokenizer.sot_sequence:
-        raise ValueError(f"This model doesn't have language tokens so it can't perform lang id")
-
-    single = mel.ndim == 2
-    if single:
-        mel = mel.unsqueeze(0)
-
-    # skip encoder forward pass if already-encoded audio features were given
-    if mel.shape[-2:] != (model.dims.n_audio_ctx, model.dims.n_audio_state):
-        mel = model.encoder(mel)
-
-    # forward pass using a single token, startoftranscript
-    n_audio = mel.shape[0]
-    x = torch.tensor([[tokenizer.sot]] * n_audio).to(mel.device)  # [n_audio, 1]
-    logits = model.logits(x, mel)[:, 0]
-
-    # collect detected languages; suppress all non-language tokens
-    mask = torch.ones(logits.shape[-1], dtype=torch.bool)
-    mask[list(tokenizer.all_language_tokens)] = False
-    logits[:, mask] = -np.inf
-    language_tokens = logits.argmax(dim=-1)
-    language_token_probs = logits.softmax(dim=-1).cpu()
-    language_probs = [
-        {
-            c: language_token_probs[i, j].item()
-            for j, c in zip(tokenizer.all_language_tokens, tokenizer.all_language_codes)
-        }
-        for i in range(n_audio)
-    ]
-
-    if single:
-        language_tokens = language_tokens[0]
-        language_probs = language_probs[0]
-
-    return language_tokens, language_probs
-
-
-@dataclass(frozen=True)
-class DecodingOptions:
-    task: str = "transcribe"  # whether to perform X->X "transcribe" or X->English "translate"
-    language: Optional[str] = None  # language that the audio is in; uses detected language if None
-
-    # sampling-related options
-    temperature: float = 0.0
-    sample_len: Optional[int] = None  # maximum number of tokens to sample
-    best_of: Optional[int] = None     # number of independent samples to collect, when t > 0
-    beam_size: Optional[int] = None   # number of beams in beam search, when t == 0
-    patience: Optional[float] = None  # patience in beam search (https://arxiv.org/abs/2204.05424)
-
-    # options for ranking generations (either beams or best-of-N samples)
-    length_penalty: Optional[float] = None   # "alpha" in Google NMT, None defaults to length norm
-
-    # prompt, prefix, and token suppression
-    prompt: Optional[Union[str, List[int]]] = None   # text or tokens for the previous context
-    prefix: Optional[Union[str, List[int]]] = None   # text or tokens to prefix the current context
-    suppress_blank: bool = True                      # this will suppress blank outputs
-
-    # list of tokens ids (or comma-separated token ids) to suppress
-    # "-1" will suppress a set of symbols as defined in `tokenizer.non_speech_tokens()`
-    suppress_tokens: Optional[Union[str, Iterable[int]]] = "-1"
-
-    # timestamp sampling options
-    without_timestamps: bool = False              # use <|notimestamps|> to sample text tokens only
-    max_initial_timestamp: Optional[float] = 1.0  # the initial timestamp cannot be later than this
-
-    # implementation details
-    fp16: bool = True  # use fp16 for most of the calculation
-
-
-@dataclass(frozen=True)
-class DecodingResult:
-    audio_features: Tensor
-    language: str
-    language_probs: Optional[Dict[str, float]] = None
-    tokens: List[int] = field(default_factory=list)
-    text: str = ""
-    avg_logprob: float = np.nan
-    no_speech_prob: float = np.nan
-    temperature: float = np.nan
-    compression_ratio: float = np.nan
-
-
-class Inference:
-    def logits(self, tokens: Tensor, audio_features: Tensor) -> Tensor:
-        """Perform a forward pass on the decoder and return per-token logits"""
-        raise NotImplementedError
-
-    def rearrange_kv_cache(self, source_indices) -> None:
-        """Update the key-value cache according to the updated beams"""
-        raise NotImplementedError
-
-    def cleanup_caching(self) -> None:
-        """Clean up any resources or hooks after decoding is finished"""
-        pass
-
-
-class PyTorchInference(Inference):
-    def __init__(self, model: "Whisper", initial_token_length: int):
-        self.model: "Whisper" = model
-        self.initial_token_length = initial_token_length
-        self.kv_cache = {}
-        self.hooks = []
-
-    def logits(self, tokens: Tensor, audio_features: Tensor) -> Tensor:
-        if not self.kv_cache:
-            self.kv_cache, self.hooks = self.model.install_kv_cache_hooks()
-
-        if tokens.shape[-1] > self.initial_token_length:
-            # only need to use the last token except in the first forward pass
-            tokens = tokens[:, -1:]
-
-        return self.model.decoder(tokens, audio_features, kv_cache=self.kv_cache)
-
-    def cleanup_caching(self):
-        for hook in self.hooks:
-            hook.remove()
-
-        self.kv_cache = {}
-        self.hooks = []
-
-    def rearrange_kv_cache(self, source_indices):
-        for module, tensor in self.kv_cache.items():
-            # update the key/value cache to contain the selected sequences
-            self.kv_cache[module] = tensor[source_indices].detach()
-
-
-class SequenceRanker:
-    def rank(self, tokens: List[List[Tensor]], sum_logprobs: List[List[float]]) -> List[int]:
-        """
-        Given a list of groups of samples and their cumulative log probabilities,
-        return the indices of the samples in each group to select as the final result
-        """
-        raise NotImplementedError
-
-
-class MaximumLikelihoodRanker(SequenceRanker):
-    """
-    Select the sample with the highest log probabilities, penalized using either
-    a simple length normalization or Google NMT paper's length penalty
-    """
-
-    def __init__(self, length_penalty: Optional[float]):
-        self.length_penalty = length_penalty
-
-    def rank(self, tokens: List[List[Tensor]], sum_logprobs: List[List[float]]):
-        def scores(logprobs, lengths):
-            result = []
-            for logprob, length in zip(logprobs, lengths):
-                if self.length_penalty is None:
-                    penalty = length
-                else:
-                    # from the Google NMT paper
-                    penalty = ((5 + length) / 6) ** self.length_penalty
-                result.append(logprob / penalty)
-            return result
-
-        # get the sequence with the highest score
-        lengths = [[len(t) for t in s] for s in tokens]
-        return [np.argmax(scores(p, l)) for p, l in zip(sum_logprobs, lengths)]
-
-
-class TokenDecoder:
-    def reset(self):
-        """Initialize any stateful variables for decoding a new sequence"""
-
-    def update(self, tokens: Tensor, logits: Tensor, sum_logprobs: Tensor) -> Tuple[Tensor, bool]:
-        """Specify how to select the next token, based on the current trace and logits
-
-        Parameters
-        ----------
-        tokens : Tensor, shape = (n_batch, current_sequence_length)
-            all tokens in the context so far, including the prefix and sot_sequence tokens
-
-        logits : Tensor, shape = (n_batch, vocab_size)
-            per-token logits of the probability distribution at the current step
-
-        sum_logprobs : Tensor, shape = (n_batch)
-            cumulative log probabilities for each sequence
-
-        Returns
-        -------
-        tokens : Tensor, shape = (n_batch, current_sequence_length + 1)
-            the tokens, appended with the selected next token
-
-        completed : bool
-            True if all sequences has reached the end of text
-
-        """
-        raise NotImplementedError
-
-    def finalize(
-        self, tokens: Tensor, sum_logprobs: Tensor
-    ) -> Tuple[Sequence[Sequence[Tensor]], List[List[float]]]:
-        """Finalize search and return the final candidate sequences
-
-        Parameters
-        ----------
-        tokens : Tensor, shape = (n_audio, n_group, current_sequence_length)
-            all tokens in the context so far, including the prefix and sot_sequence
-
-        sum_logprobs : Tensor, shape = (n_audio, n_group)
-            cumulative log probabilities for each sequence
-
-        Returns
-        -------
-        tokens : Sequence[Sequence[Tensor]], length = n_audio
-            sequence of Tensors containing candidate token sequences, for each audio input
-
-        sum_logprobs : List[List[float]], length = n_audio
-            sequence of cumulative log probabilities corresponding to the above
-
-        """
-        raise NotImplementedError
-
-
-class GreedyDecoder(TokenDecoder):
-    def __init__(self, temperature: float, eot: int):
-        self.temperature = temperature
-        self.eot = eot
-
-    def update(self, tokens: Tensor, logits: Tensor, sum_logprobs: Tensor) -> Tuple[Tensor, bool]:
-        temperature = self.temperature
-        if temperature == 0:
-            next_tokens = logits.argmax(dim=-1)
-        else:
-            next_tokens = Categorical(logits=logits / temperature).sample()
-
-        logprobs = F.log_softmax(logits.float(), dim=-1)
-        current_logprobs = logprobs[torch.arange(logprobs.shape[0]), next_tokens]
-        sum_logprobs += current_logprobs * (tokens[:, -1] != self.eot)
-
-        next_tokens[tokens[:, -1] == self.eot] = self.eot
-        tokens = torch.cat([tokens, next_tokens[:, None]], dim=-1)
-
-        completed = (tokens[:, -1] == self.eot).all()
-        return tokens, completed
-
-    def finalize(self, tokens: Tensor, sum_logprobs: Tensor):
-        # make sure each sequence has at least one EOT token at the end
-        tokens = F.pad(tokens, (0, 1), value=self.eot)
-        return tokens, sum_logprobs.tolist()
-
-
-class BeamSearchDecoder(TokenDecoder):
-    def __init__(self, beam_size: int, eot: int, inference: Inference, patience: Optional[float] = None):
-        self.beam_size = beam_size
-        self.eot = eot
-        self.inference = inference
-        self.patience = patience or 1.0
-        self.max_candidates: int = round(beam_size * self.patience)
-        self.finished_sequences = None
-
-        assert self.max_candidates > 0, f"Invalid beam size ({beam_size}) or patience ({patience})"
-
-    def reset(self):
-        self.finished_sequences = None
-
-    def update(self, tokens: Tensor, logits: Tensor, sum_logprobs: Tensor) -> Tuple[Tensor, bool]:
-        if tokens.shape[0] % self.beam_size != 0:
-            raise ValueError(f"{tokens.shape}[0] % {self.beam_size} != 0")
-
-        n_audio = tokens.shape[0] // self.beam_size
-        if self.finished_sequences is None:  # for the first update
-            self.finished_sequences = [{} for _ in range(n_audio)]
-
-        logprobs = F.log_softmax(logits.float(), dim=-1)
-        next_tokens, source_indices, finished_sequences = [], [], []
-        for i in range(n_audio):
-            scores, sources, finished = {}, {}, {}
-
-            # STEP 1: calculate the cumulative log probabilities for possible candidates
-            for j in range(self.beam_size):
-                idx = i * self.beam_size + j
-                prefix = tokens[idx].tolist()
-                for logprob, token in zip(*logprobs[idx].topk(self.beam_size + 1)):
-                    new_logprob = (sum_logprobs[idx] + logprob).item()
-                    sequence = tuple(prefix + [token.item()])
-                    scores[sequence] = new_logprob
-                    sources[sequence] = idx
-
-            # STEP 2: rank the candidates and keep the top beam_size sequences for each audio
-            saved = 0
-            for sequence in sorted(scores, key=scores.get, reverse=True):
-                if sequence[-1] == self.eot:
-                    finished[sequence] = scores[sequence]
-                else:
-                    sum_logprobs[len(next_tokens)] = scores[sequence]
-                    next_tokens.append(sequence)
-                    source_indices.append(sources[sequence])
-
-                    saved += 1
-                    if saved == self.beam_size:
-                        break
-
-            finished_sequences.append(finished)
-
-        tokens = torch.tensor(next_tokens, device=tokens.device)
-        self.inference.rearrange_kv_cache(source_indices)
-
-        # add newly finished sequences to self.finished_sequences
-        assert len(self.finished_sequences) == len(finished_sequences)
-        for previously_finished, newly_finished in zip(self.finished_sequences, finished_sequences):
-            for seq in sorted(newly_finished, key=newly_finished.get, reverse=True):
-                if len(previously_finished) >= self.max_candidates:
-                    break  # the candidate list is full
-                previously_finished[seq] = newly_finished[seq]
-
-        # mark as completed if all audio has enough number of samples
-        completed = all(
-            len(sequences) >= self.max_candidates for sequences in self.finished_sequences
-        )
-        return tokens, completed
-
-    def finalize(self, preceding_tokens: Tensor, sum_logprobs: Tensor):
-        # collect all finished sequences, including patience, and add unfinished ones if not enough
-        sum_logprobs = sum_logprobs.cpu()
-        for i, sequences in enumerate(self.finished_sequences):
-            if len(sequences) < self.beam_size:  # when not enough sequences are finished
-                for j in list(np.argsort(sum_logprobs[i]))[::-1]:
-                    sequence = preceding_tokens[i, j].tolist() + [self.eot]
-                    sequences[tuple(sequence)] = sum_logprobs[i][j].item()
-                    if len(sequences) >= self.beam_size:
-                        break
-
-        tokens: List[List[Tensor]] = [
-            [torch.tensor(seq) for seq in sequences.keys()] for sequences in self.finished_sequences
-        ]
-        sum_logprobs: List[List[float]] = [
-            list(sequences.values()) for sequences in self.finished_sequences
-        ]
-        return tokens, sum_logprobs
-
-
-class LogitFilter:
-    def apply(self, logits: Tensor, tokens: Tensor) -> None:
-        """Apply any filtering or masking to logits in-place
-
-        Parameters
-        ----------
-        logits : Tensor, shape = (n_batch, vocab_size)
-            per-token logits of the probability distribution at the current step
-
-        tokens : Tensor, shape = (n_batch, current_sequence_length)
-            all tokens in the context so far, including the prefix and sot_sequence tokens
-
-        """
-        raise NotImplementedError
-
-
-class SuppressBlank(LogitFilter):
-    def __init__(self, tokenizer: Tokenizer, sample_begin: int):
-        self.tokenizer = tokenizer
-        self.sample_begin = sample_begin
-
-    def apply(self, logits: Tensor, tokens: Tensor):
-        if tokens.shape[1] == self.sample_begin:
-            logits[:, self.tokenizer.encode(" ") + [self.tokenizer.eot]] = -np.inf
-
-
-class SuppressTokens(LogitFilter):
-    def __init__(self, suppress_tokens: Sequence[int]):
-        self.suppress_tokens = list(suppress_tokens)
-
-    def apply(self, logits: Tensor, tokens: Tensor):
-        logits[:, self.suppress_tokens] = -np.inf
-
-
-class ApplyTimestampRules(LogitFilter):
-    def __init__(
-        self, tokenizer: Tokenizer, sample_begin: int, max_initial_timestamp_index: Optional[int]
-    ):
-        self.tokenizer = tokenizer
-        self.sample_begin = sample_begin
-        self.max_initial_timestamp_index = max_initial_timestamp_index
-
-    def apply(self, logits: Tensor, tokens: Tensor):
-        # suppress <|notimestamps|> which is handled by without_timestamps
-        if self.tokenizer.no_timestamps is not None:
-            logits[:, self.tokenizer.no_timestamps] = -np.inf
-
-        # timestamps have to appear in pairs, except directly before EOT; mask logits accordingly
-        for k in range(tokens.shape[0]):
-            seq = [t for t in tokens[k, self.sample_begin :].tolist()]
-            last_was_timestamp = len(seq) >= 1 and seq[-1] >= self.tokenizer.timestamp_begin
-            penultimate_was_timestamp = len(seq) < 2 or seq[-2] >= self.tokenizer.timestamp_begin
-
-            if last_was_timestamp:
-                if penultimate_was_timestamp:  # has to be non-timestamp
-                    logits[k, self.tokenizer.timestamp_begin :] = -np.inf
-                else:  # cannot be normal text tokens
-                    logits[k, : self.tokenizer.eot] = -np.inf
-
-        if tokens.shape[1] == self.sample_begin:
-            # suppress generating non-timestamp tokens at the beginning
-            logits[:, : self.tokenizer.timestamp_begin] = -np.inf
-
-            # apply the `max_initial_timestamp` option
-            if self.max_initial_timestamp_index is not None:
-                last_allowed = self.tokenizer.timestamp_begin + self.max_initial_timestamp_index
-                logits[:, last_allowed + 1 :] = -np.inf
-
-        # if sum of probability over timestamps is above any other token, sample timestamp
-        logprobs = F.log_softmax(logits.float(), dim=-1)
-        for k in range(tokens.shape[0]):
-            timestamp_logprob = logprobs[k, self.tokenizer.timestamp_begin :].logsumexp(dim=-1)
-            max_text_token_logprob = logprobs[k, : self.tokenizer.timestamp_begin].max()
-            if timestamp_logprob > max_text_token_logprob:
-                logits[k, : self.tokenizer.timestamp_begin] = -np.inf
-
-
-class DecodingTask:
-    inference: Inference
-    sequence_ranker: SequenceRanker
-    decoder: TokenDecoder
-    logit_filters: List[LogitFilter]
-
-    def __init__(self, model: "Whisper", options: DecodingOptions):
-        self.model = model
-
-        language = options.language or "en"
-        tokenizer = get_tokenizer(model.is_multilingual, language=language, task=options.task)
-        self.tokenizer: Tokenizer = tokenizer
-        self.options: DecodingOptions = self._verify_options(options)
-
-        self.n_group: int = options.beam_size or options.best_of or 1
-        self.n_ctx: int = model.dims.n_text_ctx
-        self.sample_len: int = options.sample_len or model.dims.n_text_ctx // 2
-
-        self.sot_sequence: Tuple[int] = tokenizer.sot_sequence
-        if self.options.without_timestamps:
-            self.sot_sequence = tokenizer.sot_sequence_including_notimestamps
-
-        self.initial_tokens: Tuple[int] = self._get_initial_tokens()
-        self.sample_begin: int = len(self.initial_tokens)
-        self.sot_index: int = self.initial_tokens.index(tokenizer.sot)
-
-        # inference: implements the forward pass through the decoder, including kv caching
-        self.inference = PyTorchInference(model, len(self.initial_tokens))
-
-        # sequence ranker: implements how to rank a group of sampled sequences
-        self.sequence_ranker = MaximumLikelihoodRanker(options.length_penalty)
-
-        # decoder: implements how to select the next tokens, given the autoregressive distribution
-        if options.beam_size is not None:
-            self.decoder = BeamSearchDecoder(
-                options.beam_size, tokenizer.eot, self.inference, options.patience
-            )
-        else:
-            self.decoder = GreedyDecoder(options.temperature, tokenizer.eot)
-
-        # logit filters: applies various rules to suppress or penalize certain tokens
-        self.logit_filters = []
-        if self.options.suppress_blank:
-            self.logit_filters.append(SuppressBlank(self.tokenizer, self.sample_begin))
-        if self.options.suppress_tokens:
-            self.logit_filters.append(SuppressTokens(self._get_suppress_tokens()))
-        if not options.without_timestamps:
-            precision = CHUNK_LENGTH / model.dims.n_audio_ctx  # usually 0.02 seconds
-            max_initial_timestamp_index = None
-            if options.max_initial_timestamp:
-                max_initial_timestamp_index = round(self.options.max_initial_timestamp / precision)
-            self.logit_filters.append(
-                ApplyTimestampRules(tokenizer, self.sample_begin, max_initial_timestamp_index)
-            )
-
-    def _verify_options(self, options: DecodingOptions) -> DecodingOptions:
-        if options.beam_size is not None and options.best_of is not None:
-            raise ValueError("beam_size and best_of can't be given together")
-        if options.temperature == 0:
-            if options.best_of is not None:
-                raise ValueError("best_of with greedy sampling (T=0) is not compatible")
-        if options.patience is not None and options.beam_size is None:
-            raise ValueError("patience requires beam_size to be given")
-        if options.length_penalty is not None and not (0 <= options.length_penalty <= 1):
-            raise ValueError("length_penalty (alpha) should be a value between 0 and 1")
-
-        return options
-
-    def _get_initial_tokens(self) -> Tuple[int]:
-        tokens = list(self.sot_sequence)
-        prefix = self.options.prefix
-        prompt = self.options.prompt
-
-        if prefix:
-            prefix_tokens = (
-                self.tokenizer.encode(" " + prefix.strip()) if isinstance(prefix, str) else prefix
-            )
-            if self.sample_len is not None:
-                max_prefix_len = self.n_ctx // 2 - self.sample_len
-                prefix_tokens = prefix_tokens[-max_prefix_len:]
-            tokens = tokens + prefix_tokens
-
-        if prompt:
-            prompt_tokens = (
-                self.tokenizer.encode(" " + prompt.strip()) if isinstance(prompt, str) else prompt
-            )
-            tokens = [self.tokenizer.sot_prev] + prompt_tokens[-(self.n_ctx // 2 - 1) :] + tokens
-
-        return tuple(tokens)
-
-    def _get_suppress_tokens(self) -> Tuple[int]:
-        suppress_tokens = self.options.suppress_tokens
-
-        if isinstance(suppress_tokens, str):
-            suppress_tokens = [int(t) for t in suppress_tokens.split(",")]
-
-        if -1 in suppress_tokens:
-            suppress_tokens = [t for t in suppress_tokens if t >= 0]
-            suppress_tokens.extend(self.tokenizer.non_speech_tokens)
-        elif suppress_tokens is None or len(suppress_tokens) == 0:
-            suppress_tokens = []  # interpret empty string as an empty list
-        else:
-            assert isinstance(suppress_tokens, list), "suppress_tokens must be a list"
-
-        suppress_tokens.extend(
-            [self.tokenizer.sot, self.tokenizer.sot_prev, self.tokenizer.sot_lm]
-        )
-        if self.tokenizer.no_speech is not None:
-            # no-speech probability is collected separately
-            suppress_tokens.append(self.tokenizer.no_speech)
-
-        return tuple(sorted(set(suppress_tokens)))
-
-    def _get_audio_features(self, mel: Tensor):
-        if self.options.fp16:
-            mel = mel.half()
-
-        if mel.shape[-2:] == (self.model.dims.n_audio_ctx, self.model.dims.n_audio_state):
-            # encoded audio features are given; skip audio encoding
-            audio_features = mel
-        else:
-            audio_features = self.model.encoder(mel)
-
-        if audio_features.dtype != (torch.float16 if self.options.fp16 else torch.float32):
-            return TypeError(f"audio_features has an incorrect dtype: {audio_features.dtype}")
-
-        return audio_features
-
-    def _detect_language(self, audio_features: Tensor, tokens: Tensor):
-        languages = [self.options.language] * audio_features.shape[0]
-        lang_probs = None
-
-        if self.options.language is None or self.options.task == "lang_id":
-            lang_tokens, lang_probs = self.model.detect_language(audio_features, self.tokenizer)
-            languages = [max(probs, key=probs.get) for probs in lang_probs]
-            if self.options.language is None:
-                tokens[:, self.sot_index + 1] = lang_tokens  # write language tokens
-
-        return languages, lang_probs
-
-    def _main_loop(self, audio_features: Tensor, tokens: Tensor):
-        assert audio_features.shape[0] == tokens.shape[0]
-        n_batch = tokens.shape[0]
-        sum_logprobs: Tensor = torch.zeros(n_batch, device=audio_features.device)
-        no_speech_probs = [np.nan] * n_batch
-
-        try:
-            for i in range(self.sample_len):
-                logits = self.inference.logits(tokens, audio_features)
-
-                if i == 0 and self.tokenizer.no_speech is not None:  # save no_speech_probs
-                    probs_at_sot = logits[:, self.sot_index].float().softmax(dim=-1)
-                    no_speech_probs = probs_at_sot[:, self.tokenizer.no_speech].tolist()
-
-                # now we need to consider the logits at the last token only
-                logits = logits[:, -1]
-
-                # apply the logit filters, e.g. for suppressing or applying penalty to
-                for logit_filter in self.logit_filters:
-                    logit_filter.apply(logits, tokens)
-
-                # expand the tokens tensor with the selected next tokens
-                tokens, completed = self.decoder.update(tokens, logits, sum_logprobs)
-
-                if completed or tokens.shape[-1] > self.n_ctx:
-                    break
-        finally:
-            self.inference.cleanup_caching()
-
-        return tokens, sum_logprobs, no_speech_probs
-
-    @torch.no_grad()
-    def run(self, mel: Tensor) -> List[DecodingResult]:
-        self.decoder.reset()
-        tokenizer: Tokenizer = self.tokenizer
-        n_audio: int = mel.shape[0]
-
-        audio_features: Tensor = self._get_audio_features(mel)  # encoder forward pass
-        tokens: Tensor = torch.tensor([self.initial_tokens]).repeat(n_audio, 1)
-
-        # detect language if requested, overwriting the language token
-        languages, language_probs = self._detect_language(audio_features, tokens)
-        if self.options.task == "lang_id":
-            return [
-                DecodingResult(audio_features=features, language=language, language_probs=probs)
-                for features, language, probs in zip(audio_features, languages, language_probs)
-            ]
-
-        # repeat the audio & text tensors by the group size, for beam search or best-of-n sampling
-        audio_features = audio_features.repeat_interleave(self.n_group, dim=0)
-        tokens = tokens.repeat_interleave(self.n_group, dim=0).to(audio_features.device)
-
-        # call the main sampling loop
-        tokens, sum_logprobs, no_speech_probs = self._main_loop(audio_features, tokens)
-
-        # reshape the tensors to have (n_audio, n_group) as the first two dimensions
-        audio_features = audio_features[:: self.n_group]
-        no_speech_probs = no_speech_probs[:: self.n_group]
-        assert audio_features.shape[0] == len(no_speech_probs) == n_audio
-
-        tokens = tokens.reshape(n_audio, self.n_group, -1)
-        sum_logprobs = sum_logprobs.reshape(n_audio, self.n_group)
-
-        # get the final candidates for each group, and slice between the first sampled token and EOT
-        tokens, sum_logprobs = self.decoder.finalize(tokens, sum_logprobs)
-        tokens: List[List[Tensor]] = [
-            [t[self.sample_begin : (t == tokenizer.eot).nonzero()[0, 0]] for t in s] for s in tokens
-        ]
-
-        # select the top-ranked sample in each group
-        selected = self.sequence_ranker.rank(tokens, sum_logprobs)
-        tokens: List[List[int]] = [t[i].tolist() for i, t in zip(selected, tokens)]
-        texts: List[str] = [tokenizer.decode(t).strip() for t in tokens]
-
-        sum_logprobs: List[float] = [lp[i] for i, lp in zip(selected, sum_logprobs)]
-        avg_logprobs: List[float] = [lp / (len(t) + 1) for t, lp in zip(tokens, sum_logprobs)]
-
-        fields = (texts, languages, tokens, audio_features, avg_logprobs, no_speech_probs)
-        if len(set(map(len, fields))) != 1:
-            raise RuntimeError(f"inconsistent result lengths: {list(map(len, fields))}")
-
-        return [
-            DecodingResult(
-                audio_features=features,
-                language=language,
-                tokens=tokens,
-                text=text,
-                avg_logprob=avg_logprob,
-                no_speech_prob=no_speech_prob,
-                temperature=self.options.temperature,
-                compression_ratio=compression_ratio(text),
-            )
-            for text, language, tokens, features, avg_logprob, no_speech_prob in zip(*fields)
-        ]
-
-
-@torch.no_grad()
-def decode(model: "Whisper", mel: Tensor, options: DecodingOptions = DecodingOptions()) -> Union[DecodingResult, List[DecodingResult]]:
-    """
-    Performs decoding of 30-second audio segment(s), provided as Mel spectrogram(s).
-
-    Parameters
-    ----------
-    model: Whisper
-        the Whisper model instance
-
-    mel: torch.Tensor, shape = (80, 3000) or (*, 80, 3000)
-        A tensor containing the Mel spectrogram(s)
-
-    options: DecodingOptions
-        A dataclass that contains all necessary options for decoding 30-second segments
-
-    Returns
-    -------
-    result: Union[DecodingResult, List[DecodingResult]]
-        The result(s) of decoding contained in `DecodingResult` dataclass instance(s)
-    """
-    single = mel.ndim == 2
-    if single:
-        mel = mel.unsqueeze(0)
-
-    result = DecodingTask(model, options).run(mel)
-    
-    if single:
-        result = result[0]
-
-    return result

whisperx/diarize.py

@@ -1,56 +1,86 @@
 import numpy as np
 import pandas as pd
+from pyannote.audio import Pipeline
+from typing import Optional, Union
+import torch
 
-def assign_word_speakers(diarize_df, result_segments, fill_nearest=False):
- 
-    for seg in result_segments:
-        wdf = seg['word-segments']
-        if len(wdf['start'].dropna()) == 0:
-            wdf['start'] = seg['start']
-            wdf['end'] = seg['end']
-        speakers = []
-        for wdx, wrow in wdf.iterrows():
-            if not np.isnan(wrow['start']):
-                diarize_df['intersection'] = np.minimum(diarize_df['end'], wrow['end']) - np.maximum(diarize_df['start'], wrow['start'])
-                diarize_df['union'] = np.maximum(diarize_df['end'], wrow['end']) - np.minimum(diarize_df['start'], wrow['start'])
-                # remove no hit
-                if not fill_nearest:
-                    dia_tmp = diarize_df[diarize_df['intersection'] > 0]
-                else:
-                    dia_tmp = diarize_df
-                if len(dia_tmp) == 0:
-                    speaker = None
-                else:
-                    speaker = dia_tmp.sort_values("intersection", ascending=False).iloc[0][2]
-            else:
-                speaker = None
-            speakers.append(speaker)
-        seg['word-segments']['speaker'] = speakers
-        seg["speaker"] = pd.Series(speakers).value_counts().index[0]
+from whisperx.audio import load_audio, SAMPLE_RATE
+from whisperx.types import TranscriptionResult, AlignedTranscriptionResult
 
-    # create word level segments for .srt
-    word_seg = []
-    for seg in result_segments:
-        wseg = pd.DataFrame(seg["word-segments"])
-        for wdx, wrow in wseg.iterrows():
-            if wrow["start"] is not None:
-                speaker = wrow['speaker']
-                if speaker is None or speaker == np.nan:
-                    speaker = "UNKNOWN"
-                word_seg.append(
-                    {
-                        "start": wrow["start"],
-                        "end": wrow["end"],
-                        "text": f"[{speaker}]: " + seg["text"][int(wrow["segment-text-start"]):int(wrow["segment-text-end"])]
-                    }
-                )
 
-    # TODO: create segments but split words on new speaker
+class DiarizationPipeline:
+    def __init__(
+        self,
+        model_name=None,
+        use_auth_token=None,
+        device: Optional[Union[str, torch.device]] = "cpu",
+    ):
+        if isinstance(device, str):
+            device = torch.device(device)
+        model_config = model_name or "pyannote/speaker-diarization-3.1"
+        self.model = Pipeline.from_pretrained(model_config, use_auth_token=use_auth_token).to(device)
+
+    def __call__(
+        self,
+        audio: Union[str, np.ndarray],
+        num_speakers: Optional[int] = None,
+        min_speakers: Optional[int] = None,
+        max_speakers: Optional[int] = None,
+    ):
+        if isinstance(audio, str):
+            audio = load_audio(audio)
+        audio_data = {
+            'waveform': torch.from_numpy(audio[None, :]),
+            'sample_rate': SAMPLE_RATE
+        }
+        segments = self.model(audio_data, num_speakers = num_speakers, min_speakers=min_speakers, max_speakers=max_speakers)
+        diarize_df = pd.DataFrame(segments.itertracks(yield_label=True), columns=['segment', 'label', 'speaker'])
+        diarize_df['start'] = diarize_df['segment'].apply(lambda x: x.start)
+        diarize_df['end'] = diarize_df['segment'].apply(lambda x: x.end)
+        return diarize_df
+
+
+def assign_word_speakers(
+    diarize_df: pd.DataFrame,
+    transcript_result: Union[AlignedTranscriptionResult, TranscriptionResult],
+    fill_nearest=False,
+) -> dict:
+    transcript_segments = transcript_result["segments"]
+    for seg in transcript_segments:
+        # assign speaker to segment (if any)
+        diarize_df['intersection'] = np.minimum(diarize_df['end'], seg['end']) - np.maximum(diarize_df['start'], seg['start'])
+        diarize_df['union'] = np.maximum(diarize_df['end'], seg['end']) - np.minimum(diarize_df['start'], seg['start'])
+        # remove no hit, otherwise we look for closest (even negative intersection...)
+        if not fill_nearest:
+            dia_tmp = diarize_df[diarize_df['intersection'] > 0]
+        else:
+            dia_tmp = diarize_df
+        if len(dia_tmp) > 0:
+            # sum over speakers
+            speaker = dia_tmp.groupby("speaker")["intersection"].sum().sort_values(ascending=False).index[0]
+            seg["speaker"] = speaker
+        
+        # assign speaker to words
+        if 'words' in seg:
+            for word in seg['words']:
+                if 'start' in word:
+                    diarize_df['intersection'] = np.minimum(diarize_df['end'], word['end']) - np.maximum(diarize_df['start'], word['start'])
+                    diarize_df['union'] = np.maximum(diarize_df['end'], word['end']) - np.minimum(diarize_df['start'], word['start'])
+                    # remove no hit
+                    if not fill_nearest:
+                        dia_tmp = diarize_df[diarize_df['intersection'] > 0]
+                    else:
+                        dia_tmp = diarize_df
+                    if len(dia_tmp) > 0:
+                        # sum over speakers
+                        speaker = dia_tmp.groupby("speaker")["intersection"].sum().sort_values(ascending=False).index[0]
+                        word["speaker"] = speaker
+        
+    return transcript_result            
 
-    return result_segments, word_seg
 
 class Segment:
-    def __init__(self, start, end, speaker=None):
+    def __init__(self, start:int, end:int, speaker:Optional[str]=None):
         self.start = start
         self.end = end
         self.speaker = speaker

whisperx/model.py

@@ -1,268 +0,0 @@
-from dataclasses import dataclass
-from typing import Dict
-from typing import Iterable, Optional
-
-import numpy as np
-import torch
-import torch.nn.functional as F
-from torch import Tensor
-from torch import nn
-
-from .transcribe import transcribe as transcribe_function
-from .decoding import detect_language as detect_language_function, decode as decode_function
-
-
-@dataclass
-class ModelDimensions:
-    n_mels: int
-    n_audio_ctx: int
-    n_audio_state: int
-    n_audio_head: int
-    n_audio_layer: int
-    n_vocab: int
-    n_text_ctx: int
-    n_text_state: int
-    n_text_head: int
-    n_text_layer: int
-
-
-class LayerNorm(nn.LayerNorm):
-    def forward(self, x: Tensor) -> Tensor:
-        return super().forward(x.float()).type(x.dtype)
-
-
-class Linear(nn.Linear):
-    def forward(self, x: Tensor) -> Tensor:
-        return F.linear(
-            x, self.weight.to(x.dtype), None if self.bias is None else self.bias.to(x.dtype)
-        )
-
-
-class Conv1d(nn.Conv1d):
-    def _conv_forward(self, x: Tensor, weight: Tensor, bias: Optional[Tensor]) -> Tensor:
-        return super()._conv_forward(
-            x, weight.to(x.dtype), None if bias is None else bias.to(x.dtype)
-        )
-
-
-def sinusoids(length, channels, max_timescale=10000):
-    """Returns sinusoids for positional embedding"""
-    assert channels % 2 == 0
-    log_timescale_increment = np.log(max_timescale) / (channels // 2 - 1)
-    inv_timescales = torch.exp(-log_timescale_increment * torch.arange(channels // 2))
-    scaled_time = torch.arange(length)[:, np.newaxis] * inv_timescales[np.newaxis, :]
-    return torch.cat([torch.sin(scaled_time), torch.cos(scaled_time)], dim=1)
-
-
-class MultiHeadAttention(nn.Module):
-    def __init__(self, n_state: int, n_head: int):
-        super().__init__()
-        self.n_head = n_head
-        self.query = Linear(n_state, n_state)
-        self.key = Linear(n_state, n_state, bias=False)
-        self.value = Linear(n_state, n_state)
-        self.out = Linear(n_state, n_state)
-
-    def forward(
-        self,
-        x: Tensor,
-        xa: Optional[Tensor] = None,
-        mask: Optional[Tensor] = None,
-        kv_cache: Optional[dict] = None,
-    ):
-        q = self.query(x)
-
-        if kv_cache is None or xa is None or self.key not in kv_cache:
-            # hooks, if installed (i.e. kv_cache is not None), will prepend the cached kv tensors;
-            # otherwise, perform key/value projections for self- or cross-attention as usual.
-            k = self.key(x if xa is None else xa)
-            v = self.value(x if xa is None else xa)
-        else:
-            # for cross-attention, calculate keys and values once and reuse in subsequent calls.
-            k = kv_cache[self.key]
-            v = kv_cache[self.value]
-
-        wv, qk = self.qkv_attention(q, k, v, mask)
-        return self.out(wv), qk
-
-    def qkv_attention(self, q: Tensor, k: Tensor, v: Tensor, mask: Optional[Tensor] = None):
-        n_batch, n_ctx, n_state = q.shape
-        scale = (n_state // self.n_head) ** -0.25
-        q = q.view(*q.shape[:2], self.n_head, -1).permute(0, 2, 1, 3) * scale
-        k = k.view(*k.shape[:2], self.n_head, -1).permute(0, 2, 3, 1) * scale
-        v = v.view(*v.shape[:2], self.n_head, -1).permute(0, 2, 1, 3)
-
-        qk = q @ k
-        if mask is not None:
-            qk = qk + mask[:n_ctx, :n_ctx]
-        qk = qk.float()
-
-        w = F.softmax(qk, dim=-1).to(q.dtype)
-        return (w @ v).permute(0, 2, 1, 3).flatten(start_dim=2), qk.detach()
-
-
-class ResidualAttentionBlock(nn.Module):
-    def __init__(self, n_state: int, n_head: int, cross_attention: bool = False):
-        super().__init__()
-
-        self.attn = MultiHeadAttention(n_state, n_head)
-        self.attn_ln = LayerNorm(n_state)
-
-        self.cross_attn = MultiHeadAttention(n_state, n_head) if cross_attention else None
-        self.cross_attn_ln = LayerNorm(n_state) if cross_attention else None
-
-        n_mlp = n_state * 4
-        self.mlp = nn.Sequential(Linear(n_state, n_mlp), nn.GELU(), Linear(n_mlp, n_state))
-        self.mlp_ln = LayerNorm(n_state)
-
-    def forward(
-        self,
-        x: Tensor,
-        xa: Optional[Tensor] = None,
-        mask: Optional[Tensor] = None,
-        kv_cache: Optional[dict] = None,
-    ):
-        x = x + self.attn(self.attn_ln(x), mask=mask, kv_cache=kv_cache)[0]
-        if self.cross_attn:
-            x = x + self.cross_attn(self.cross_attn_ln(x), xa, kv_cache=kv_cache)[0]
-        x = x + self.mlp(self.mlp_ln(x))
-        return x
-
-
-class AudioEncoder(nn.Module):
-    def __init__(self, n_mels: int, n_ctx: int, n_state: int, n_head: int, n_layer: int):
-        super().__init__()
-        self.conv1 = Conv1d(n_mels, n_state, kernel_size=3, padding=1)
-        self.conv2 = Conv1d(n_state, n_state, kernel_size=3, stride=2, padding=1)
-        self.register_buffer("positional_embedding", sinusoids(n_ctx, n_state))
-
-        self.blocks: Iterable[ResidualAttentionBlock] = nn.ModuleList(
-            [ResidualAttentionBlock(n_state, n_head) for _ in range(n_layer)]
-        )
-        self.ln_post = LayerNorm(n_state)
-
-    def forward(self, x: Tensor):
-        """
-        x : torch.Tensor, shape = (batch_size, n_mels, n_ctx)
-            the mel spectrogram of the audio
-        """
-        x = F.gelu(self.conv1(x))
-        x = F.gelu(self.conv2(x))
-        x = x.permute(0, 2, 1)
-
-        assert x.shape[1:] == self.positional_embedding.shape, "incorrect audio shape"
-        x = (x + self.positional_embedding).to(x.dtype)
-
-        for block in self.blocks:
-            x = block(x)
-
-        x = self.ln_post(x)
-        return x
-
-
-class TextDecoder(nn.Module):
-    def __init__(self, n_vocab: int, n_ctx: int, n_state: int, n_head: int, n_layer: int):
-        super().__init__()
-
-        self.token_embedding = nn.Embedding(n_vocab, n_state)
-        self.positional_embedding = nn.Parameter(torch.empty(n_ctx, n_state))
-
-        self.blocks: Iterable[ResidualAttentionBlock] = nn.ModuleList(
-            [ResidualAttentionBlock(n_state, n_head, cross_attention=True) for _ in range(n_layer)]
-        )
-        self.ln = LayerNorm(n_state)
-
-        mask = torch.empty(n_ctx, n_ctx).fill_(-np.inf).triu_(1)
-        self.register_buffer("mask", mask, persistent=False)
-
-    def forward(self, x: Tensor, xa: Tensor, kv_cache: Optional[dict] = None):
-        """
-        x : torch.LongTensor, shape = (batch_size, <= n_ctx)
-            the text tokens
-        xa : torch.Tensor, shape = (batch_size, n_mels, n_audio_ctx)
-            the encoded audio features to be attended on
-        """
-        offset = next(iter(kv_cache.values())).shape[1] if kv_cache else 0
-        x = self.token_embedding(x) + self.positional_embedding[offset : offset + x.shape[-1]]
-        x = x.to(xa.dtype)
-
-        for block in self.blocks:
-            x = block(x, xa, mask=self.mask, kv_cache=kv_cache)
-
-        x = self.ln(x)
-        logits = (x @ torch.transpose(self.token_embedding.weight.to(x.dtype), 0, 1)).float()
-
-        return logits
-
-
-class Whisper(nn.Module):
-    def __init__(self, dims: ModelDimensions):
-        super().__init__()
-        self.dims = dims
-        self.encoder = AudioEncoder(
-            self.dims.n_mels,
-            self.dims.n_audio_ctx,
-            self.dims.n_audio_state,
-            self.dims.n_audio_head,
-            self.dims.n_audio_layer,
-        )
-        self.decoder = TextDecoder(
-            self.dims.n_vocab,
-            self.dims.n_text_ctx,
-            self.dims.n_text_state,
-            self.dims.n_text_head,
-            self.dims.n_text_layer,
-        )
-
-    def embed_audio(self, mel: torch.Tensor):
-        return self.encoder(mel)
-
-    def logits(self, tokens: torch.Tensor, audio_features: torch.Tensor):
-        return self.decoder(tokens, audio_features)
-
-    def forward(self, mel: torch.Tensor, tokens: torch.Tensor) -> Dict[str, torch.Tensor]:
-        return self.decoder(tokens, self.encoder(mel))
-
-    @property
-    def device(self):
-        return next(self.parameters()).device
-
-    @property
-    def is_multilingual(self):
-        return self.dims.n_vocab == 51865
-
-    def install_kv_cache_hooks(self, cache: Optional[dict] = None):
-        """
-        The `MultiHeadAttention` module optionally accepts `kv_cache` which stores the key and value
-        tensors calculated for the previous positions. This method returns a dictionary that stores
-        all caches, and the necessary hooks for the key and value projection modules that save the
-        intermediate tensors to be reused during later calculations.
-
-        Returns
-        -------
-        cache : Dict[nn.Module, torch.Tensor]
-            A dictionary object mapping the key/value projection modules to its cache
-        hooks : List[RemovableHandle]
-            List of PyTorch RemovableHandle objects to stop the hooks to be called
-        """
-        cache = {**cache} if cache is not None else {}
-        hooks = []
-
-        def save_to_cache(module, _, output):
-            if module not in cache or output.shape[1] > self.decoder.positional_embedding.shape[0]:
-                cache[module] = output  # save as-is, for the first token or cross attention
-            else:
-                cache[module] = torch.cat([cache[module], output], dim=1).detach()
-            return cache[module]
-
-        def install_hooks(layer: nn.Module):
-            if isinstance(layer, MultiHeadAttention):
-                hooks.append(layer.key.register_forward_hook(save_to_cache))
-                hooks.append(layer.value.register_forward_hook(save_to_cache))
-
-        self.decoder.apply(install_hooks)
-        return cache, hooks
-
-    detect_language = detect_language_function
-    transcribe = transcribe_function
-    decode = decode_function

whisperx/normalizers/__init__.py

@@ -1,2 +0,0 @@
-from .basic import BasicTextNormalizer
-from .english import EnglishTextNormalizer

whisperx/normalizers/basic.py

@@ -1,71 +0,0 @@
-import re
-import unicodedata
-
-import regex
-
-# non-ASCII letters that are not separated by "NFKD" normalization
-ADDITIONAL_DIACRITICS = {
-    "œ": "oe",
-    "Œ": "OE",
-    "ø": "o",
-    "Ø": "O",
-    "æ": "ae",
-    "Æ": "AE",
-    "ß": "ss",
-    "ẞ": "SS",
-    "đ": "d",
-    "Đ": "D",
-    "ð": "d",
-    "Ð": "D",
-    "þ": "th",
-    "Þ": "th",
-    "ł": "l",
-    "Ł": "L",
-}
-
-
-def remove_symbols_and_diacritics(s: str, keep=""):
-    """
-    Replace any other markers, symbols, and punctuations with a space,
-    and drop any diacritics (category 'Mn' and some manual mappings)
-    """
-    return "".join(
-        c
-        if c in keep
-        else ADDITIONAL_DIACRITICS[c]
-        if c in ADDITIONAL_DIACRITICS
-        else ""
-        if unicodedata.category(c) == "Mn"
-        else " "
-        if unicodedata.category(c)[0] in "MSP"
-        else c
-        for c in unicodedata.normalize("NFKD", s)
-    )
-
-
-def remove_symbols(s: str):
-    """
-    Replace any other markers, symbols, punctuations with a space, keeping diacritics
-    """
-    return "".join(
-        " " if unicodedata.category(c)[0] in "MSP" else c for c in unicodedata.normalize("NFKC", s)
-    )
-
-
-class BasicTextNormalizer:
-    def __init__(self, remove_diacritics: bool = False, split_letters: bool = False):
-        self.clean = remove_symbols_and_diacritics if remove_diacritics else remove_symbols
-        self.split_letters = split_letters
-
-    def __call__(self, s: str):
-        s = s.lower()
-        s = re.sub(r"[<\[][^>\]]*[>\]]", "", s)  # remove words between brackets
-        s = re.sub(r"\(([^)]+?)\)", "", s)  # remove words between parenthesis
-        s = self.clean(s).lower()
-
-        if self.split_letters:
-            s = " ".join(regex.findall(r"\X", s, regex.U))
-
-        s = re.sub(r"\s+", " ", s)  # replace any successive whitespace characters with a space
-
-        return s

whisperx/normalizers/english.py

@@ -1,543 +0,0 @@
-import json
-import os
-import re
-from fractions import Fraction
-from typing import Iterator, List, Match, Optional, Union
-
-from more_itertools import windowed
-
-from .basic import remove_symbols_and_diacritics
-
-
-class EnglishNumberNormalizer:
-    """
-    Convert any spelled-out numbers into arabic numbers, while handling:
-
-    - remove any commas
-    - keep the suffixes such as: `1960s`, `274th`, `32nd`, etc.
-    - spell out currency symbols after the number. e.g. `$20 million` -> `20000000 dollars`
-    - spell out `one` and `ones`
-    - interpret successive single-digit numbers as nominal: `one oh one` -> `101`
-    """
-
-    def __init__(self):
-        super().__init__()
-
-        self.zeros = {"o", "oh", "zero"}
-        self.ones = {
-            name: i
-            for i, name in enumerate(
-                [
-                    "one",
-                    "two",
-                    "three",
-                    "four",
-                    "five",
-                    "six",
-                    "seven",
-                    "eight",
-                    "nine",
-                    "ten",
-                    "eleven",
-                    "twelve",
-                    "thirteen",
-                    "fourteen",
-                    "fifteen",
-                    "sixteen",
-                    "seventeen",
-                    "eighteen",
-                    "nineteen",
-                ],
-                start=1,
-            )
-        }
-        self.ones_plural = {
-            "sixes" if name == "six" else name + "s": (value, "s")
-            for name, value in self.ones.items()
-        }
-        self.ones_ordinal = {
-            "zeroth": (0, "th"),
-            "first": (1, "st"),
-            "second": (2, "nd"),
-            "third": (3, "rd"),
-            "fifth": (5, "th"),
-            "twelfth": (12, "th"),
-            **{
-                name + ("h" if name.endswith("t") else "th"): (value, "th")
-                for name, value in self.ones.items()
-                if value > 3 and value != 5 and value != 12
-            },
-        }
-        self.ones_suffixed = {**self.ones_plural, **self.ones_ordinal}
-
-        self.tens = {
-            "twenty": 20,
-            "thirty": 30,
-            "forty": 40,
-            "fifty": 50,
-            "sixty": 60,
-            "seventy": 70,
-            "eighty": 80,
-            "ninety": 90,
-        }
-        self.tens_plural = {
-            name.replace("y", "ies"): (value, "s") for name, value in self.tens.items()
-        }
-        self.tens_ordinal = {
-            name.replace("y", "ieth"): (value, "th") for name, value in self.tens.items()
-        }
-        self.tens_suffixed = {**self.tens_plural, **self.tens_ordinal}
-
-        self.multipliers = {
-            "hundred": 100,
-            "thousand": 1_000,
-            "million": 1_000_000,
-            "billion": 1_000_000_000,
-            "trillion": 1_000_000_000_000,
-            "quadrillion": 1_000_000_000_000_000,
-            "quintillion": 1_000_000_000_000_000_000,
-            "sextillion": 1_000_000_000_000_000_000_000,
-            "septillion": 1_000_000_000_000_000_000_000_000,
-            "octillion": 1_000_000_000_000_000_000_000_000_000,
-            "nonillion": 1_000_000_000_000_000_000_000_000_000_000,
-            "decillion": 1_000_000_000_000_000_000_000_000_000_000_000,
-        }
-        self.multipliers_plural = {
-            name + "s": (value, "s") for name, value in self.multipliers.items()
-        }
-        self.multipliers_ordinal = {
-            name + "th": (value, "th") for name, value in self.multipliers.items()
-        }
-        self.multipliers_suffixed = {**self.multipliers_plural, **self.multipliers_ordinal}
-        self.decimals = {*self.ones, *self.tens, *self.zeros}
-
-        self.preceding_prefixers = {
-            "minus": "-",
-            "negative": "-",
-            "plus": "+",
-            "positive": "+",
-        }
-        self.following_prefixers = {
-            "pound": "£",
-            "pounds": "£",
-            "euro": "€",
-            "euros": "€",
-            "dollar": "$",
-            "dollars": "$",
-            "cent": "¢",
-            "cents": "¢",
-        }
-        self.prefixes = set(
-            list(self.preceding_prefixers.values()) + list(self.following_prefixers.values())
-        )
-        self.suffixers = {
-            "per": {"cent": "%"},
-            "percent": "%",
-        }
-        self.specials = {"and", "double", "triple", "point"}
-
-        self.words = set(
-            [
-                key
-                for mapping in [
-                    self.zeros,
-                    self.ones,
-                    self.ones_suffixed,
-                    self.tens,
-                    self.tens_suffixed,
-                    self.multipliers,
-                    self.multipliers_suffixed,
-                    self.preceding_prefixers,
-                    self.following_prefixers,
-                    self.suffixers,
-                    self.specials,
-                ]
-                for key in mapping
-            ]
-        )
-        self.literal_words = {"one", "ones"}
-
-    def process_words(self, words: List[str]) -> Iterator[str]:
-        prefix: Optional[str] = None
-        value: Optional[Union[str, int]] = None
-        skip = False
-
-        def to_fraction(s: str):
-            try:
-                return Fraction(s)
-            except ValueError:
-                return None
-
-        def output(result: Union[str, int]):
-            nonlocal prefix, value
-            result = str(result)
-            if prefix is not None:
-                result = prefix + result
-            value = None
-            prefix = None
-            return result
-
-        if len(words) == 0:
-            return
-
-        for prev, current, next in windowed([None] + words + [None], 3):
-            if skip:
-                skip = False
-                continue
-
-            next_is_numeric = next is not None and re.match(r"^\d+(\.\d+)?$", next)
-            has_prefix = current[0] in self.prefixes
-            current_without_prefix = current[1:] if has_prefix else current
-            if re.match(r"^\d+(\.\d+)?$", current_without_prefix):
-                # arabic numbers (potentially with signs and fractions)
-                f = to_fraction(current_without_prefix)
-                assert f is not None
-                if value is not None:
-                    if isinstance(value, str) and value.endswith("."):
-                        # concatenate decimals / ip address components
-                        value = str(value) + str(current)
-                        continue
-                    else:
-                        yield output(value)
-
-                prefix = current[0] if has_prefix else prefix
-                if f.denominator == 1:
-                    value = f.numerator  # store integers as int
-                else:
-                    value = current_without_prefix
-            elif current not in self.words:
-                # non-numeric words
-                if value is not None:
-                    yield output(value)
-                yield output(current)
-            elif current in self.zeros:
-                value = str(value or "") + "0"
-            elif current in self.ones:
-                ones = self.ones[current]
-
-                if value is None:
-                    value = ones
-                elif isinstance(value, str) or prev in self.ones:
-                    if prev in self.tens and ones < 10:  # replace the last zero with the digit
-                        assert value[-1] == "0"
-                        value = value[:-1] + str(ones)
-                    else:
-                        value = str(value) + str(ones)
-                elif ones < 10:
-                    if value % 10 == 0:
-                        value += ones
-                    else:
-                        value = str(value) + str(ones)
-                else:  # eleven to nineteen
-                    if value % 100 == 0:
-                        value += ones
-                    else:
-                        value = str(value) + str(ones)
-            elif current in self.ones_suffixed:
-                # ordinal or cardinal; yield the number right away
-                ones, suffix = self.ones_suffixed[current]
-                if value is None:
-                    yield output(str(ones) + suffix)
-                elif isinstance(value, str) or prev in self.ones:
-                    if prev in self.tens and ones < 10:
-                        assert value[-1] == "0"
-                        yield output(value[:-1] + str(ones) + suffix)
-                    else:
-                        yield output(str(value) + str(ones) + suffix)
-                elif ones < 10:
-                    if value % 10 == 0:
-                        yield output(str(value + ones) + suffix)
-                    else:
-                        yield output(str(value) + str(ones) + suffix)
-                else:  # eleven to nineteen
-                    if value % 100 == 0:
-                        yield output(str(value + ones) + suffix)
-                    else:
-                        yield output(str(value) + str(ones) + suffix)
-                value = None
-            elif current in self.tens:
-                tens = self.tens[current]
-                if value is None:
-                    value = tens
-                elif isinstance(value, str):
-                    value = str(value) + str(tens)
-                else:
-                    if value % 100 == 0:
-                        value += tens
-                    else:
-                        value = str(value) + str(tens)
-            elif current in self.tens_suffixed:
-                # ordinal or cardinal; yield the number right away
-                tens, suffix = self.tens_suffixed[current]
-                if value is None:
-                    yield output(str(tens) + suffix)
-                elif isinstance(value, str):
-                    yield output(str(value) + str(tens) + suffix)
-                else:
-                    if value % 100 == 0:
-                        yield output(str(value + tens) + suffix)
-                    else:
-                        yield output(str(value) + str(tens) + suffix)
-            elif current in self.multipliers:
-                multiplier = self.multipliers[current]
-                if value is None:
-                    value = multiplier
-                elif isinstance(value, str) or value == 0:
-                    f = to_fraction(value)
-                    p = f * multiplier if f is not None else None
-                    if f is not None and p.denominator == 1:
-                        value = p.numerator
-                    else:
-                        yield output(value)
-                        value = multiplier
-                else:
-                    before = value // 1000 * 1000
-                    residual = value % 1000
-                    value = before + residual * multiplier
-            elif current in self.multipliers_suffixed:
-                multiplier, suffix = self.multipliers_suffixed[current]
-                if value is None:
-                    yield output(str(multiplier) + suffix)
-                elif isinstance(value, str):
-                    f = to_fraction(value)
-                    p = f * multiplier if f is not None else None
-                    if f is not None and p.denominator == 1:
-                        yield output(str(p.numerator) + suffix)
-                    else:
-                        yield output(value)
-                        yield output(str(multiplier) + suffix)
-                else:  # int
-                    before = value // 1000 * 1000
-                    residual = value % 1000
-                    value = before + residual * multiplier
-                    yield output(str(value) + suffix)
-                value = None
-            elif current in self.preceding_prefixers:
-                # apply prefix (positive, minus, etc.) if it precedes a number
-                if value is not None:
-                    yield output(value)
-
-                if next in self.words or next_is_numeric:
-                    prefix = self.preceding_prefixers[current]
-                else:
-                    yield output(current)
-            elif current in self.following_prefixers:
-                # apply prefix (dollars, cents, etc.) only after a number
-                if value is not None:
-                    prefix = self.following_prefixers[current]
-                    yield output(value)
-                else:
-                    yield output(current)
-            elif current in self.suffixers:
-                # apply suffix symbols (percent -> '%')
-                if value is not None:
-                    suffix = self.suffixers[current]
-                    if isinstance(suffix, dict):
-                        if next in suffix:
-                            yield output(str(value) + suffix[next])
-                            skip = True
-                        else:
-                            yield output(value)
-                            yield output(current)
-                    else:
-                        yield output(str(value) + suffix)
-                else:
-                    yield output(current)
-            elif current in self.specials:
-                if next not in self.words and not next_is_numeric:
-                    # apply special handling only if the next word can be numeric
-                    if value is not None:
-                        yield output(value)
-                    yield output(current)
-                elif current == "and":
-                    # ignore "and" after hundreds, thousands, etc.
-                    if prev not in self.multipliers:
-                        if value is not None:
-                            yield output(value)
-                        yield output(current)
-                elif current == "double" or current == "triple":
-                    if next in self.ones or next in self.zeros:
-                        repeats = 2 if current == "double" else 3
-                        ones = self.ones.get(next, 0)
-                        value = str(value or "") + str(ones) * repeats
-                        skip = True
-                    else:
-                        if value is not None:
-                            yield output(value)
-                        yield output(current)
-                elif current == "point":
-                    if next in self.decimals or next_is_numeric:
-                        value = str(value or "") + "."
-                else:
-                    # should all have been covered at this point
-                    raise ValueError(f"Unexpected token: {current}")
-            else:
-                # all should have been covered at this point
-                raise ValueError(f"Unexpected token: {current}")
-
-        if value is not None:
-            yield output(value)
-
-    def preprocess(self, s: str):
-        # replace "<number> and a half" with "<number> point five"
-        results = []
-
-        segments = re.split(r"\band\s+a\s+half\b", s)
-        for i, segment in enumerate(segments):
-            if len(segment.strip()) == 0:
-                continue
-            if i == len(segments) - 1:
-                results.append(segment)
-            else:
-                results.append(segment)
-                last_word = segment.rsplit(maxsplit=2)[-1]
-                if last_word in self.decimals or last_word in self.multipliers:
-                    results.append("point five")
-                else:
-                    results.append("and a half")
-
-        s = " ".join(results)
-
-        # put a space at number/letter boundary
-        s = re.sub(r"([a-z])([0-9])", r"\1 \2", s)
-        s = re.sub(r"([0-9])([a-z])", r"\1 \2", s)
-
-        # but remove spaces which could be a suffix
-        s = re.sub(r"([0-9])\s+(st|nd|rd|th|s)\b", r"\1\2", s)
-
-        return s
-
-    def postprocess(self, s: str):
-        def combine_cents(m: Match):
-            try:
-                currency = m.group(1)
-                integer = m.group(2)
-                cents = int(m.group(3))
-                return f"{currency}{integer}.{cents:02d}"
-            except ValueError:
-                return m.string
-
-        def extract_cents(m: Match):
-            try:
-                return f"¢{int(m.group(1))}"
-            except ValueError:
-                return m.string
-
-        # apply currency postprocessing; "$2 and ¢7" -> "$2.07"
-        s = re.sub(r"([€£$])([0-9]+) (?:and )?¢([0-9]{1,2})\b", combine_cents, s)
-        s = re.sub(r"[€£$]0.([0-9]{1,2})\b", extract_cents, s)
-
-        # write "one(s)" instead of "1(s)", just for the readability
-        s = re.sub(r"\b1(s?)\b", r"one\1", s)
-
-        return s
-
-    def __call__(self, s: str):
-        s = self.preprocess(s)
-        s = " ".join(word for word in self.process_words(s.split()) if word is not None)
-        s = self.postprocess(s)
-
-        return s
-
-
-class EnglishSpellingNormalizer:
-    """
-    Applies British-American spelling mappings as listed in [1].
-
-    [1] https://www.tysto.com/uk-us-spelling-list.html
-    """
-
-    def __init__(self):
-        mapping_path = os.path.join(os.path.dirname(__file__), "english.json")
-        self.mapping = json.load(open(mapping_path))
-
-    def __call__(self, s: str):
-        return " ".join(self.mapping.get(word, word) for word in s.split())
-
-
-class EnglishTextNormalizer:
-    def __init__(self):
-        self.ignore_patterns = r"\b(hmm|mm|mhm|mmm|uh|um)\b"
-        self.replacers = {
-            # common contractions
-            r"\bwon't\b": "will not",
-            r"\bcan't\b": "can not",
-            r"\blet's\b": "let us",
-            r"\bain't\b": "aint",
-            r"\by'all\b": "you all",
-            r"\bwanna\b": "want to",
-            r"\bgotta\b": "got to",
-            r"\bgonna\b": "going to",
-            r"\bi'ma\b": "i am going to",
-            r"\bimma\b": "i am going to",
-            r"\bwoulda\b": "would have",
-            r"\bcoulda\b": "could have",
-            r"\bshoulda\b": "should have",
-            r"\bma'am\b": "madam",
-            # contractions in titles/prefixes
-            r"\bmr\b": "mister ",
-            r"\bmrs\b": "missus ",
-            r"\bst\b": "saint ",
-            r"\bdr\b": "doctor ",
-            r"\bprof\b": "professor ",
-            r"\bcapt\b": "captain ",
-            r"\bgov\b": "governor ",
-            r"\bald\b": "alderman ",
-            r"\bgen\b": "general ",
-            r"\bsen\b": "senator ",
-            r"\brep\b": "representative ",
-            r"\bpres\b": "president ",
-            r"\brev\b": "reverend ",
-            r"\bhon\b": "honorable ",
-            r"\basst\b": "assistant ",
-            r"\bassoc\b": "associate ",
-            r"\blt\b": "lieutenant ",
-            r"\bcol\b": "colonel ",
-            r"\bjr\b": "junior ",
-            r"\bsr\b": "senior ",
-            r"\besq\b": "esquire ",
-            # prefect tenses, ideally it should be any past participles, but it's harder..
-            r"'d been\b": " had been",
-            r"'s been\b": " has been",
-            r"'d gone\b": " had gone",
-            r"'s gone\b": " has gone",
-            r"'d done\b": " had done",  # "'s done" is ambiguous
-            r"'s got\b": " has got",
-            # general contractions
-            r"n't\b": " not",
-            r"'re\b": " are",
-            r"'s\b": " is",
-            r"'d\b": " would",
-            r"'ll\b": " will",
-            r"'t\b": " not",
-            r"'ve\b": " have",
-            r"'m\b": " am",
-        }
-        self.standardize_numbers = EnglishNumberNormalizer()
-        self.standardize_spellings = EnglishSpellingNormalizer()
-
-    def __call__(self, s: str):
-        s = s.lower()
-
-        s = re.sub(r"[<\[][^>\]]*[>\]]", "", s)  # remove words between brackets
-        s = re.sub(r"\(([^)]+?)\)", "", s)  # remove words between parenthesis
-        s = re.sub(self.ignore_patterns, "", s)
-        s = re.sub(r"\s+'", "'", s)  # standardize when there's a space before an apostrophe
-
-        for pattern, replacement in self.replacers.items():
-            s = re.sub(pattern, replacement, s)
-
-        s = re.sub(r"(\d),(\d)", r"\1\2", s)  # remove commas between digits
-        s = re.sub(r"\.([^0-9]|$)", r" \1", s)  # remove periods not followed by numbers
-        s = remove_symbols_and_diacritics(s, keep=".%$¢€£")  # keep some symbols for numerics
-
-        s = self.standardize_numbers(s)
-        s = self.standardize_spellings(s)
-
-        # now remove prefix/suffix symbols that are not preceded/followed by numbers
-        s = re.sub(r"[.$¢€£]([^0-9])", r" \1", s)
-        s = re.sub(r"([^0-9])%", r"\1 ", s)
-
-        s = re.sub(r"\s+", " ", s)  # replace any successive whitespace characters with a space
-
-        return s

whisperx/tokenizer.py

@@ -1,331 +0,0 @@
-import os
-from dataclasses import dataclass
-from functools import lru_cache
-from typing import List, Optional, Tuple, Union
-
-import numpy as np
-import torch
-from transformers import GPT2TokenizerFast
-
-LANGUAGES = {
-    "en": "english",
-    "zh": "chinese",
-    "de": "german",
-    "es": "spanish",
-    "ru": "russian",
-    "ko": "korean",
-    "fr": "french",
-    "ja": "japanese",
-    "pt": "portuguese",
-    "tr": "turkish",
-    "pl": "polish",
-    "ca": "catalan",
-    "nl": "dutch",
-    "ar": "arabic",
-    "sv": "swedish",
-    "it": "italian",
-    "id": "indonesian",
-    "hi": "hindi",
-    "fi": "finnish",
-    "vi": "vietnamese",
-    "he": "hebrew",
-    "uk": "ukrainian",
-    "el": "greek",
-    "ms": "malay",
-    "cs": "czech",
-    "ro": "romanian",
-    "da": "danish",
-    "hu": "hungarian",
-    "ta": "tamil",
-    "no": "norwegian",
-    "th": "thai",
-    "ur": "urdu",
-    "hr": "croatian",
-    "bg": "bulgarian",
-    "lt": "lithuanian",
-    "la": "latin",
-    "mi": "maori",
-    "ml": "malayalam",
-    "cy": "welsh",
-    "sk": "slovak",
-    "te": "telugu",
-    "fa": "persian",
-    "lv": "latvian",
-    "bn": "bengali",
-    "sr": "serbian",
-    "az": "azerbaijani",
-    "sl": "slovenian",
-    "kn": "kannada",
-    "et": "estonian",
-    "mk": "macedonian",
-    "br": "breton",
-    "eu": "basque",
-    "is": "icelandic",
-    "hy": "armenian",
-    "ne": "nepali",
-    "mn": "mongolian",
-    "bs": "bosnian",
-    "kk": "kazakh",
-    "sq": "albanian",
-    "sw": "swahili",
-    "gl": "galician",
-    "mr": "marathi",
-    "pa": "punjabi",
-    "si": "sinhala",
-    "km": "khmer",
-    "sn": "shona",
-    "yo": "yoruba",
-    "so": "somali",
-    "af": "afrikaans",
-    "oc": "occitan",
-    "ka": "georgian",
-    "be": "belarusian",
-    "tg": "tajik",
-    "sd": "sindhi",
-    "gu": "gujarati",
-    "am": "amharic",
-    "yi": "yiddish",
-    "lo": "lao",
-    "uz": "uzbek",
-    "fo": "faroese",
-    "ht": "haitian creole",
-    "ps": "pashto",
-    "tk": "turkmen",
-    "nn": "nynorsk",
-    "mt": "maltese",
-    "sa": "sanskrit",
-    "lb": "luxembourgish",
-    "my": "myanmar",
-    "bo": "tibetan",
-    "tl": "tagalog",
-    "mg": "malagasy",
-    "as": "assamese",
-    "tt": "tatar",
-    "haw": "hawaiian",
-    "ln": "lingala",
-    "ha": "hausa",
-    "ba": "bashkir",
-    "jw": "javanese",
-    "su": "sundanese",
-}
-
-# language code lookup by name, with a few language aliases
-TO_LANGUAGE_CODE = {
-    **{language: code for code, language in LANGUAGES.items()},
-    "burmese": "my",
-    "valencian": "ca",
-    "flemish": "nl",
-    "haitian": "ht",
-    "letzeburgesch": "lb",
-    "pushto": "ps",
-    "panjabi": "pa",
-    "moldavian": "ro",
-    "moldovan": "ro",
-    "sinhalese": "si",
-    "castilian": "es",
-}
-
-
-@dataclass(frozen=True)
-class Tokenizer:
-    """A thin wrapper around `GPT2TokenizerFast` providing quick access to special tokens"""
-
-    tokenizer: "GPT2TokenizerFast"
-    language: Optional[str]
-    sot_sequence: Tuple[int]
-
-    def encode(self, text, **kwargs):
-        return self.tokenizer.encode(text, **kwargs)
-
-    def decode(self, token_ids: Union[int, List[int], np.ndarray, torch.Tensor], **kwargs):
-        return self.tokenizer.decode(token_ids, **kwargs)
-
-    def decode_with_timestamps(self, tokens) -> str:
-        """
-        Timestamp tokens are above the special tokens' id range and are ignored by `decode()`.
-        This method decodes given tokens with timestamps tokens annotated, e.g. "<|1.08|>".
-        """
-        outputs = [[]]
-        for token in tokens:
-            if token >= self.timestamp_begin:
-                timestamp = f"<|{(token - self.timestamp_begin) * 0.02:.2f}|>"
-                outputs.append(timestamp)
-                outputs.append([])
-            else:
-                outputs[-1].append(token)
-        outputs = [s if isinstance(s, str) else self.tokenizer.decode(s) for s in outputs]
-        return "".join(outputs)
-
-    @property
-    @lru_cache()
-    def eot(self) -> int:
-        return self.tokenizer.eos_token_id
-
-    @property
-    @lru_cache()
-    def sot(self) -> int:
-        return self._get_single_token_id("<|startoftranscript|>")
-
-    @property
-    @lru_cache()
-    def sot_lm(self) -> int:
-        return self._get_single_token_id("<|startoflm|>")
-
-    @property
-    @lru_cache()
-    def sot_prev(self) -> int:
-        return self._get_single_token_id("<|startofprev|>")
-
-    @property
-    @lru_cache()
-    def no_speech(self) -> int:
-        return self._get_single_token_id("<|nospeech|>")
-
-    @property
-    @lru_cache()
-    def no_timestamps(self) -> int:
-        return self._get_single_token_id("<|notimestamps|>")
-
-    @property
-    @lru_cache()
-    def timestamp_begin(self) -> int:
-        return self.tokenizer.all_special_ids[-1] + 1
-
-    @property
-    @lru_cache()
-    def language_token(self) -> int:
-        """Returns the token id corresponding to the value of the `language` field"""
-        if self.language is None:
-            raise ValueError(f"This tokenizer does not have language token configured")
-
-        additional_tokens = dict(
-            zip(
-                self.tokenizer.additional_special_tokens,
-                self.tokenizer.additional_special_tokens_ids,
-            )
-        )
-        candidate = f"<|{self.language}|>"
-        if candidate in additional_tokens:
-            return additional_tokens[candidate]
-
-        raise KeyError(f"Language {self.language} not found in tokenizer.")
-
-    @property
-    @lru_cache()
-    def all_language_tokens(self) -> Tuple[int]:
-        result = []
-        for token, token_id in zip(
-            self.tokenizer.additional_special_tokens,
-            self.tokenizer.additional_special_tokens_ids,
-        ):
-            if token.strip("<|>") in LANGUAGES:
-                result.append(token_id)
-        return tuple(result)
-
-    @property
-    @lru_cache()
-    def all_language_codes(self) -> Tuple[str]:
-        return tuple(self.decode([l]).strip("<|>") for l in self.all_language_tokens)
-
-    @property
-    @lru_cache()
-    def sot_sequence_including_notimestamps(self) -> Tuple[int]:
-        return tuple(list(self.sot_sequence) + [self.no_timestamps])
-
-    @property
-    @lru_cache()
-    def non_speech_tokens(self) -> Tuple[int]:
-        """
-        Returns the list of tokens to suppress in order to avoid any speaker tags or non-speech
-        annotations, to prevent sampling texts that are not actually spoken in the audio, e.g.
-
-        - ♪♪♪
-        - ( SPEAKING FOREIGN LANGUAGE )
-        - [DAVID] Hey there,
-
-        keeping basic punctuations like commas, periods, question marks, exclamation points, etc.
-        """
-        symbols = list("\"#()*+/:;<=>@[\\]^_`{|}~「」『』")
-        symbols += "<< >> <<< >>> -- --- -( -[ (' (\" (( )) ((( ))) [[ ]] {{ }} ♪♪ ♪♪♪".split()
-
-        # symbols that may be a single token or multiple tokens depending on the tokenizer.
-        # In case they're multiple tokens, suppress the first token, which is safe because:
-        # These are between U+2640 and U+267F miscellaneous symbols that are okay to suppress
-        # in generations, and in the 3-byte UTF-8 representation they share the first two bytes.
-        miscellaneous = set("♩♪♫♬♭♮♯")
-        assert all(0x2640 <= ord(c) <= 0x267F for c in miscellaneous)
-
-        # allow hyphens "-" and single quotes "'" between words, but not at the beginning of a word
-        result = {self.tokenizer.encode(" -")[0], self.tokenizer.encode(" '")[0]}
-        for symbol in symbols + list(miscellaneous):
-            for tokens in [self.tokenizer.encode(symbol), self.tokenizer.encode(" " + symbol)]:
-                if len(tokens) == 1 or symbol in miscellaneous:
-                    result.add(tokens[0])
-
-        return tuple(sorted(result))
-
-    def _get_single_token_id(self, text) -> int:
-        tokens = self.tokenizer.encode(text)
-        assert len(tokens) == 1, f"{text} is not encoded as a single token"
-        return tokens[0]
-
-
-@lru_cache(maxsize=None)
-def build_tokenizer(name: str = "gpt2"):
-    os.environ["TOKENIZERS_PARALLELISM"] = "false"
-    path = os.path.join(os.path.dirname(__file__), "assets", name)
-    tokenizer = GPT2TokenizerFast.from_pretrained(path)
-
-    specials = [
-        "<|startoftranscript|>",
-        *[f"<|{lang}|>" for lang in LANGUAGES.keys()],
-        "<|translate|>",
-        "<|transcribe|>",
-        "<|startoflm|>",
-        "<|startofprev|>",
-        "<|nospeech|>",
-        "<|notimestamps|>",
-    ]
-
-    tokenizer.add_special_tokens(dict(additional_special_tokens=specials))
-    return tokenizer
-
-
-@lru_cache(maxsize=None)
-def get_tokenizer(
-    multilingual: bool,
-    *,
-    task: Optional[str] = None,  # Literal["transcribe", "translate", None]
-    language: Optional[str] = None,
-) -> Tokenizer:
-    if language is not None:
-        language = language.lower()
-        if language not in LANGUAGES:
-            if language in TO_LANGUAGE_CODE:
-                language = TO_LANGUAGE_CODE[language]
-            else:
-                raise ValueError(f"Unsupported language: {language}")
-
-    if multilingual:
-        tokenizer_name = "multilingual"
-        task = task or "transcribe"
-        language = language or "en"
-    else:
-        tokenizer_name = "gpt2"
-        task = None
-        language = None
-
-    tokenizer = build_tokenizer(name=tokenizer_name)
-    all_special_ids: List[int] = tokenizer.all_special_ids
-    sot: int = all_special_ids[1]
-    translate: int = all_special_ids[-6]
-    transcribe: int = all_special_ids[-5]
-
-    langs = tuple(LANGUAGES.keys())
-    sot_sequence = [sot]
-    if language is not None:
-        sot_sequence.append(sot + 1 + langs.index(language))
-    if task is not None:
-        sot_sequence.append(transcribe if task == "transcribe" else translate)
-
-    return Tokenizer(tokenizer=tokenizer, language=language, sot_sequence=tuple(sot_sequence))

whisperx/transcribe.py

@@ -1,763 +1,220 @@
 import argparse
+import gc
 import os
 import warnings
-from typing import List, Optional, Tuple, Union, Iterator, TYPE_CHECKING
 
 import numpy as np
 import torch
-import tqdm
-from .audio import SAMPLE_RATE, N_FRAMES, HOP_LENGTH, CHUNK_LENGTH, pad_or_trim, log_mel_spectrogram, load_audio
-from .alignment import load_align_model, align, get_trellis, backtrack, merge_repeats, merge_words
-from .decoding import DecodingOptions, DecodingResult
-from .diarize import assign_word_speakers, Segment
-from .tokenizer import LANGUAGES, TO_LANGUAGE_CODE, get_tokenizer
-from .utils import exact_div, format_timestamp, optional_int, optional_float, str2bool, interpolate_nans, write_txt, write_vtt, write_srt, write_ass, write_tsv
-from .vad import Binarize
-import pandas as pd
 
-if TYPE_CHECKING:
-    from .model import Whisper
+from whisperx.alignment import align, load_align_model
+from whisperx.asr import load_model
+from whisperx.audio import load_audio
+from whisperx.diarize import DiarizationPipeline, assign_word_speakers
+from whisperx.types import AlignedTranscriptionResult, TranscriptionResult
+from whisperx.utils import LANGUAGES, TO_LANGUAGE_CODE, get_writer
 
 
+def transcribe_task(args: dict, parser: argparse.ArgumentParser):
+    """Transcription task to be called from CLI.
 
-def transcribe(
-    model: "Whisper",
-    audio: Union[str, np.ndarray, torch.Tensor],
-    *,
-    verbose: Optional[bool] = None,
-    temperature: Union[float, Tuple[float, ...]] = (0.0, 0.2, 0.4, 0.6, 0.8, 1.0),
-    compression_ratio_threshold: Optional[float] = 2.4,
-    logprob_threshold: Optional[float] = -1.0,
-    no_speech_threshold: Optional[float] = 0.6,
-    condition_on_previous_text: bool = False, # turn off by default due to errors it causes
-    mel: np.ndarray = None,
-    **decode_options,
-):
+    Args:
+        args: Dictionary of command-line arguments.
+        parser: argparse.ArgumentParser object.
     """
-    Transcribe an audio file using Whisper
+    # fmt: off
 
-    Parameters
-    ----------
-    model: Whisper
-        The Whisper model instance
-
-    audio: Union[str, np.ndarray, torch.Tensor]
-        The path to the audio file to open, or the audio waveform
-
-    verbose: bool
-        Whether to display the text being decoded to the console. If True, displays all the details,
-        If False, displays minimal details. If None, does not display anything
-
-    temperature: Union[float, Tuple[float, ...]]
-        Temperature for sampling. It can be a tuple of temperatures, which will be successfully used
-        upon failures according to either `compression_ratio_threshold` or `logprob_threshold`.
-
-    compression_ratio_threshold: float
-        If the gzip compression ratio is above this value, treat as failed
-
-    logprob_threshold: float
-        If the average log probability over sampled tokens is below this value, treat as failed
-
-    no_speech_threshold: float
-        If the no_speech probability is higher than this value AND the average log probability
-        over sampled tokens is below `logprob_threshold`, consider the segment as silent
-
-    condition_on_previous_text: bool
-        if True, the previous output of the model is provided as a prompt for the next window;
-        disabling may make the text inconsistent across windows, but the model becomes less prone to
-        getting stuck in a failure loop, such as repetition looping or timestamps going out of sync.
-
-    decode_options: dict
-        Keyword arguments to construct `DecodingOptions` instances
-
-    Returns
-    -------
-    A dictionary containing the resulting text ("text") and segment-level details ("segments"), and
-    the spoken language ("language"), which is detected when `decode_options["language"]` is None.
-    """
-    dtype = torch.float16 if decode_options.get("fp16", True) else torch.float32
-    if model.device == torch.device("cpu"):
-        if torch.cuda.is_available():
-            warnings.warn("Performing inference on CPU when CUDA is available")
-        if dtype == torch.float16:
-            warnings.warn("FP16 is not supported on CPU; using FP32 instead")
-            dtype = torch.float32
-
-    if dtype == torch.float32:
-        decode_options["fp16"] = False
-
-    if mel is None:
-        mel = log_mel_spectrogram(audio)
-
-    if decode_options.get("language", None) is None:
-        if not model.is_multilingual:
-            decode_options["language"] = "en"
-        else:
-            if verbose:
-                print("Detecting language using up to the first 30 seconds. Use `--language` to specify the language")
-            segment = pad_or_trim(mel, N_FRAMES).to(model.device).to(dtype)
-            _, probs = model.detect_language(segment)
-            decode_options["language"] = max(probs, key=probs.get)
-            if verbose is not None:
-                print(f"Detected language: {LANGUAGES[decode_options['language']].title()}")
-
-    language = decode_options["language"]
-    task = decode_options.get("task", "transcribe")
-    tokenizer = get_tokenizer(model.is_multilingual, language=language, task=task)
-
-    def decode_with_fallback(segment: torch.Tensor) -> DecodingResult:
-        temperatures = [temperature] if isinstance(temperature, (int, float)) else temperature
-        decode_result = None
-
-        for t in temperatures:
-            kwargs = {**decode_options}
-            if t > 0:
-                # disable beam_size and patience when t > 0
-                kwargs.pop("beam_size", None)
-                kwargs.pop("patience", None)
-            else:
-                # disable best_of when t == 0
-                kwargs.pop("best_of", None)
-
-            options = DecodingOptions(**kwargs, temperature=t)
-            decode_result = model.decode(segment, options)
-
-            needs_fallback = False
-            if compression_ratio_threshold is not None and decode_result.compression_ratio > compression_ratio_threshold:
-                needs_fallback = True  # too repetitive
-            if logprob_threshold is not None and decode_result.avg_logprob < logprob_threshold:
-                needs_fallback = True  # average log probability is too low
-
-            if not needs_fallback:
-                break
-
-        return decode_result
-
-    seek = 0
-    input_stride = exact_div(
-        N_FRAMES, model.dims.n_audio_ctx
-    )  # mel frames per output token: 2
-    time_precision = (
-        input_stride * HOP_LENGTH / SAMPLE_RATE
-    )  # time per output token: 0.02 (seconds)
-    all_tokens = []
-    all_segments = []
-    prompt_reset_since = 0
-
-    initial_prompt = decode_options.pop("initial_prompt", None) or []
-    if initial_prompt:
-        initial_prompt = tokenizer.encode(" " + initial_prompt.strip())
-        all_tokens.extend(initial_prompt)
-
-    def add_segment(
-        *, start: float, end: float, text_tokens: torch.Tensor, result: DecodingResult
-    ):
-        text = tokenizer.decode([token for token in text_tokens if token < tokenizer.eot])
-        if len(text.strip()) == 0:  # skip empty text output
-            return
-
-        all_segments.append(
-            {
-                "id": len(all_segments),
-                "seek": seek,
-                "start": start,
-                "end": end,
-                "text": text,
-                "tokens": text_tokens.tolist(),
-                "temperature": result.temperature,
-                "avg_logprob": result.avg_logprob,
-                "compression_ratio": result.compression_ratio,
-                "no_speech_prob": result.no_speech_prob,
-            }
-        )
-        if verbose:
-            print(f"[{format_timestamp(start)} --> {format_timestamp(end)}] {text}")
-
-    # show the progress bar when verbose is False (otherwise the transcribed text will be printed)
-    num_frames = mel.shape[-1]
-    previous_seek_value = seek
-
-    with tqdm.tqdm(total=num_frames, unit='frames', disable=verbose is not False) as pbar:
-        while seek < num_frames:
-            timestamp_offset = float(seek * HOP_LENGTH / SAMPLE_RATE)
-            segment = pad_or_trim(mel[:, seek:], N_FRAMES).to(model.device).to(dtype)
-            segment_duration = segment.shape[-1] * HOP_LENGTH / SAMPLE_RATE
-
-            decode_options["prompt"] = all_tokens[prompt_reset_since:]
-            result: DecodingResult = decode_with_fallback(segment)
-            tokens = torch.tensor(result.tokens)
-
-            if no_speech_threshold is not None:
-                # no voice activity check
-                should_skip = result.no_speech_prob > no_speech_threshold
-                if logprob_threshold is not None and result.avg_logprob > logprob_threshold:
-                    # don't skip if the logprob is high enough, despite the no_speech_prob
-                    should_skip = False
-
-                if should_skip:
-                    seek += segment.shape[-1]  # fast-forward to the next segment boundary
-                    continue
-
-            timestamp_tokens: torch.Tensor = tokens.ge(tokenizer.timestamp_begin)
-            consecutive = torch.where(timestamp_tokens[:-1] & timestamp_tokens[1:])[0].add_(1)
-            if len(consecutive) > 0:  # if the output contains two consecutive timestamp tokens
-                last_slice = 0
-                for current_slice in consecutive:
-                    sliced_tokens = tokens[last_slice:current_slice]
-                    start_timestamp_position = (
-                        sliced_tokens[0].item() - tokenizer.timestamp_begin
-                    )
-                    end_timestamp_position = (
-                        sliced_tokens[-1].item() - tokenizer.timestamp_begin
-                    )
-
-                    # clamp end-time to at least be 1 frame after start-time
-                    end_timestamp_position = max(end_timestamp_position, start_timestamp_position + time_precision)
-
-                    add_segment(
-                        start=timestamp_offset + start_timestamp_position * time_precision,
-                        end=timestamp_offset + end_timestamp_position * time_precision,
-                        text_tokens=sliced_tokens[1:-1],
-                        result=result,
-                    )
-                    last_slice = current_slice
-                last_timestamp_position = (
-                    tokens[last_slice - 1].item() - tokenizer.timestamp_begin
-                )
-                seek += last_timestamp_position * input_stride
-                all_tokens.extend(tokens[: last_slice + 1].tolist())
-            else:
-                duration = segment_duration
-                timestamps = tokens[timestamp_tokens.nonzero().flatten()]
-                if len(timestamps) > 0 and timestamps[-1].item() != tokenizer.timestamp_begin:
-                    # no consecutive timestamps but it has a timestamp; use the last one.
-                    # single timestamp at the end means no speech after the last timestamp.
-                    last_timestamp_position = timestamps[-1].item() - tokenizer.timestamp_begin
-                    duration = last_timestamp_position * time_precision
-
-                add_segment(
-                    start=timestamp_offset,
-                    end=timestamp_offset + duration,
-                    text_tokens=tokens,
-                    result=result,
-                )
-
-                seek += segment.shape[-1]
-                all_tokens.extend(tokens.tolist())
-
-            if not condition_on_previous_text or result.temperature > 0.5:
-                # do not feed the prompt tokens if a high temperature was used
-                prompt_reset_since = len(all_tokens)
-
-            # update progress bar
-            pbar.update(min(num_frames, seek) - previous_seek_value)
-            previous_seek_value = seek
-
-    return dict(text=tokenizer.decode(all_tokens[len(initial_prompt):]), segments=all_segments, language=language)
-
-
-def merge_chunks(segments, chunk_size=CHUNK_LENGTH):
-    """
-    Merge VAD segments into larger segments of approximately size ~CHUNK_LENGTH.
-    TODO: Make sure VAD segment isn't too long, otherwise it will cause OOM when input to alignment model
-    TODO: Or sliding window alignment model over long segment.
-    """
-    curr_end = 0
-    merged_segments = []
-    seg_idxs = []
-    speaker_idxs = []
-
-    assert chunk_size > 0
-    binarize = Binarize(max_duration=chunk_size)
-    segments = binarize(segments)
-    segments_list = []
-    for speech_turn in segments.get_timeline():
-        segments_list.append(Segment(speech_turn.start, speech_turn.end, "UNKNOWN"))
-
-    assert segments_list, "segments_list is empty."
-    # Make sur the starting point is the start of the segment.
-    curr_start = segments_list[0].start
-
-    for seg in segments_list:
-        if seg.end - curr_start > chunk_size and curr_end-curr_start > 0:
-            merged_segments.append({
-                "start": curr_start,
-                "end": curr_end,
-                "segments": seg_idxs,
-            })
-            curr_start = seg.start
-            seg_idxs = []
-            speaker_idxs = []
-        curr_end = seg.end
-        seg_idxs.append((seg.start, seg.end))
-        speaker_idxs.append(seg.speaker)
-    # add final
-    merged_segments.append({ 
-                "start": curr_start,
-                "end": curr_end,
-                "segments": seg_idxs,
-            })    
-    return merged_segments
-
-
-def transcribe_with_vad(
-    model: "Whisper",
-    audio: Union[str, np.ndarray, torch.Tensor],
-    vad_pipeline,
-    mel = None,
-    verbose: Optional[bool] = None,
-    **kwargs
-):
-    """
-    Transcribe per VAD segment
-    """
-
-    if mel is None:
-        mel = log_mel_spectrogram(audio)
-    
-    prev = 0
-    output = {"segments": []}
-
-    vad_segments = vad_pipeline(audio)
-    # merge segments to approx 30s inputs to make whisper most appropraite
-    vad_segments = merge_chunks(vad_segments)
-
-    for sdx, seg_t in enumerate(vad_segments):
-        if verbose:
-            print(f"~~ Transcribing VAD chunk: ({format_timestamp(seg_t['start'])} --> {format_timestamp(seg_t['end'])}) ~~")
-        seg_f_start, seg_f_end = int(seg_t["start"] * SAMPLE_RATE / HOP_LENGTH), int(seg_t["end"] * SAMPLE_RATE / HOP_LENGTH)
-        local_f_start, local_f_end = seg_f_start - prev, seg_f_end - prev
-        mel = mel[:, local_f_start:] # seek forward
-        prev = seg_f_start
-        local_mel = mel[:, :local_f_end-local_f_start]
-        result = transcribe(model, audio, mel=local_mel, verbose=verbose, **kwargs)
-        seg_t["text"] = result["text"]
-        output["segments"].append(
-            {
-                "start": seg_t["start"],
-                "end": seg_t["end"],
-                "language": result["language"],
-                "text": result["text"],
-                "seg-text": [x["text"] for x in result["segments"]],
-                "seg-start": [x["start"] for x in result["segments"]],
-                "seg-end": [x["end"] for x in result["segments"]],
-                }
-            )
-
-    output["language"] = output["segments"][0]["language"]
-
-    return output
-
-
-def transcribe_with_vad_parallel(
-    model: "Whisper",
-    audio: Union[str, np.ndarray, torch.Tensor],
-    vad_pipeline,
-    mel = None,
-    verbose: Optional[bool] = None,
-    batch_size = -1,
-    **kwargs
-):
-    """
-    Transcribe per VAD segment
-    """
-
-    if mel is None:
-        mel = log_mel_spectrogram(audio)
-    
-    vad_segments = vad_pipeline(audio)
-    # merge segments to approx 30s inputs to make whisper most appropraite
-    vad_segments = merge_chunks(vad_segments)
-
-    ################################
-    ### START of parallelization ###
-    ################################
-
-    # pad mel to a same length
-    start_seconds = [i['start'] for i in vad_segments]
-    end_seconds = [i['end'] for i in vad_segments]
-    duration_list = np.array(end_seconds) - np.array(start_seconds)
-    max_length = round(30 / (HOP_LENGTH / SAMPLE_RATE))
-    offset_list = np.array(start_seconds)
-    chunks = []
-
-    for start_ts, end_ts in zip(start_seconds, end_seconds):
-        start_ts = round(start_ts / (HOP_LENGTH / SAMPLE_RATE))
-        end_ts = round(end_ts / (HOP_LENGTH / SAMPLE_RATE))
-        chunk = mel[:, start_ts:end_ts]
-        chunk = torch.nn.functional.pad(chunk, (0, max_length-chunk.shape[-1]))
-        chunks.append(chunk)
-    
-    mel_chunk = torch.stack(chunks, dim=0).to(model.device)
-    # using 'decode_options1': only support single temperature decoding (no fallbacks)
-    # result_list2 = model.decode(mel_chunk, decode_options1)
-
-    # prepare DecodingOptions
-    temperatures = kwargs.pop("temperature", None)
-    compression_ratio_threshold = kwargs.pop("compression_ratio_threshold", None)
-    logprob_threshold = kwargs.pop("logprob_threshold", None)
-    no_speech_threshold = kwargs.pop("no_speech_threshold", None)
-    condition_on_previous_text = kwargs.pop("condition_on_previous_text", None)
-    initial_prompt = kwargs.pop("initial_prompt", None)
-    
-    t = 0  # TODO: does not upport temperature sweeping
-    if t > 0:
-        # disable beam_size and patience when t > 0
-        kwargs.pop("beam_size", None)
-        kwargs.pop("patience", None)
-    else:
-        # disable best_of when t == 0
-        kwargs.pop("best_of", None)
-
-    options = DecodingOptions(**kwargs, temperature=t)
-    mel_chunk_batches = torch.split(mel_chunk, split_size_or_sections=batch_size)
-    decode_result = []
-    for mel_chunk_batch in mel_chunk_batches:
-        decode_result.extend(model.decode(mel_chunk_batch, options))
-    
-    ##############################
-    ### END of parallelization ###
-    ##############################
-
-    # post processing: get segments rfom batch-decoded results
-    input_stride = exact_div(
-        N_FRAMES, model.dims.n_audio_ctx
-    )  # mel frames per output token: 2
-    language = kwargs["language"]
-    task = kwargs["task"]
-    tokenizer = get_tokenizer(model.is_multilingual, language=language, task=task)
-
-    output = post_process_results(
-        vad_segments,
-        decode_result, 
-        duration_list, 
-        offset_list,
-        input_stride,
-        language,
-        tokenizer,
-        no_speech_threshold=no_speech_threshold,
-        logprob_threshold=logprob_threshold,
-        verbose=verbose)
-    return output
-
-
-def post_process_results(
-    vad_segments,
-    result_list, 
-    duration_list, 
-    offset_list,
-    input_stride,
-    language,
-    tokenizer,
-    no_speech_threshold = None,
-    logprob_threshold = None,
-    verbose: Optional[bool] = None,
-    ):
-
-    seek = 0
-    time_precision = (
-        input_stride * HOP_LENGTH / SAMPLE_RATE
-    )  # time per output token: 0.02 (seconds)
-    all_tokens = []
-    all_segments = []
-    output = {"segments": []}
-
-    def add_segment(
-        *, start: float, end: float, text_tokens: torch.Tensor, result: DecodingResult
-    ):
-        text = tokenizer.decode([token for token in text_tokens if token < tokenizer.eot])
-        if len(text.strip()) == 0:  # skip empty text output
-            return
-
-        all_segments.append(
-            {
-                "id": len(all_segments),
-                "seek": seek,
-                "start": start,
-                "end": end,
-                "text": text,
-                "tokens": text_tokens.tolist(),
-                "temperature": result.temperature,
-                "avg_logprob": result.avg_logprob,
-                "compression_ratio": result.compression_ratio,
-                "no_speech_prob": result.no_speech_prob,
-            }
-        )
-        if verbose:
-            print(f"[{format_timestamp(start)} --> {format_timestamp(end)}] {text}")
-
-    # process the output
-    for seg_t, result, segment_duration, timestamp_offset in zip(vad_segments, result_list, duration_list, offset_list):
-        all_tokens = []
-        all_segments = []
-
-        # segment_duration = segment.shape[-1] * HOP_LENGTH / SAMPLE_RATE
-        segment_shape = int(segment_duration / (HOP_LENGTH / SAMPLE_RATE))
-        tokens = torch.tensor(result.tokens)
-
-        if no_speech_threshold is not None:
-            # no voice activity check
-            should_skip = result.no_speech_prob > no_speech_threshold
-            if logprob_threshold is not None and result.avg_logprob > logprob_threshold:
-                # don't skip if the logprob is high enough, despite the no_speech_prob
-                should_skip = False
-
-            if should_skip:
-                seek += segment_shape  # fast-forward to the next segment boundary
-                continue
-        
-        timestamp_tokens: torch.Tensor = tokens.ge(tokenizer.timestamp_begin)
-        consecutive = torch.where(timestamp_tokens[:-1] & timestamp_tokens[1:])[0].add_(1)
-        
-        if len(consecutive) > 0:  # if the output contains two consecutive timestamp tokens
-            last_slice = 0
-            for current_slice in consecutive:
-                sliced_tokens = tokens[last_slice:current_slice]
-                start_timestamp_position = (
-                    sliced_tokens[0].item() - tokenizer.timestamp_begin
-                )
-                end_timestamp_position = (
-                    sliced_tokens[-1].item() - tokenizer.timestamp_begin
-                )
-                add_segment(
-                    start=timestamp_offset + start_timestamp_position * time_precision,
-                    end=timestamp_offset + end_timestamp_position * time_precision,
-                    text_tokens=sliced_tokens[1:-1],
-                    result=result,
-                )
-                last_slice = current_slice
-            last_timestamp_position = (
-                tokens[last_slice - 1].item() - tokenizer.timestamp_begin
-            )
-            seek += last_timestamp_position * input_stride
-            all_tokens.extend(tokens[: last_slice + 1].tolist())
-        else:
-            duration = segment_duration
-            timestamps = tokens[timestamp_tokens.nonzero().flatten()]
-            if len(timestamps) > 0 and timestamps[-1].item() != tokenizer.timestamp_begin:
-                # no consecutive timestamps but it has a timestamp; use the last one.
-                # single timestamp at the end means no speech after the last timestamp.
-                last_timestamp_position = timestamps[-1].item() - tokenizer.timestamp_begin
-                duration = last_timestamp_position * time_precision
-
-            add_segment(
-                start=timestamp_offset,
-                end=timestamp_offset + duration,
-                text_tokens=tokens,
-                result=result,
-            )
-
-            seek += segment_shape
-            all_tokens.extend(tokens.tolist())
-
-        result = dict(text=tokenizer.decode(all_tokens), segments=all_segments, language=language)
-        output["segments"].append(
-            {
-                "start": seg_t["start"],
-                "end": seg_t["end"],
-                "language": result["language"],
-                "text": result["text"],
-                "seg-text": [x["text"] for x in result["segments"]],
-                "seg-start": [x["start"] for x in result["segments"]],
-                "seg-end": [x["end"] for x in result["segments"]],
-                }
-            )
-
-    output["language"] = output["segments"][0]["language"]
-
-    return output
-
-
-def cli():
-    from . import available_models
-
-    parser = argparse.ArgumentParser(formatter_class=argparse.ArgumentDefaultsHelpFormatter)
-    parser.add_argument("audio", nargs="+", type=str, help="audio file(s) to transcribe")
-    parser.add_argument("--model", default="small", choices=available_models(), help="name of the Whisper model to use")
-    parser.add_argument("--model_dir", type=str, default=None, help="the path to save model files; uses ~/.cache/whisper by default")
-    parser.add_argument("--device", default="cuda" if torch.cuda.is_available() else "cpu", help="device to use for PyTorch inference")
-    # alignment params
-    parser.add_argument("--align_model", default=None, help="Name of phoneme-level ASR model to do alignment")
-    parser.add_argument("--align_extend", default=2, type=float, help="Seconds before and after to extend the whisper segments for alignment")
-    parser.add_argument("--align_from_prev", default=True, type=bool, help="Whether to clip the alignment start time of current segment to the end time of the last aligned word of the previous segment")
-    parser.add_argument("--interpolate_method", default="nearest", choices=["nearest", "linear", "ignore"], help="For word .srt, method to assign timestamps to non-aligned words, or merge them into neighbouring.")
-    # vad params
-    parser.add_argument("--vad_filter", action="store_true", help="Whether to first perform VAD filtering to target only transcribe within VAD. Produces more accurate alignment + timestamp, requires more GPU memory & compute.")
-    parser.add_argument("--parallel_bs", default=-1, type=int, help="Enable parallel transcribing if > 1")
-    # diarization params
-    parser.add_argument("--diarize", action="store_true", help="Apply diarization to assign speaker labels to each segment/word")
-    parser.add_argument("--min_speakers", default=None, type=int)
-    parser.add_argument("--max_speakers", default=None, type=int)
-    # output save params
-    parser.add_argument("--output_dir", "-o", type=str, default=".", help="directory to save the outputs")
-    parser.add_argument("--output_type", default="all", choices=["all", "srt", "srt-word", "vtt", "txt", "tsv", "ass", "ass-char", "pickle", "vad"], help="File type for desired output save")
-
-    parser.add_argument("--verbose", type=str2bool, default=True, help="whether to print out the progress and debug messages")
-
-    parser.add_argument("--task", type=str, default="transcribe", choices=["transcribe", "translate"], help="whether to perform X->X speech recognition ('transcribe') or X->English translation ('translate')")
-    parser.add_argument("--language", type=str, default=None, choices=sorted(LANGUAGES.keys()) + sorted([k.title() for k in TO_LANGUAGE_CODE.keys()]), help="language spoken in the audio, specify None to perform language detection")
-
-    parser.add_argument("--temperature", type=float, default=0, help="temperature to use for sampling")
-    parser.add_argument("--best_of", type=optional_int, default=5, help="number of candidates when sampling with non-zero temperature")
-    parser.add_argument("--beam_size", type=optional_int, default=5, help="number of beams in beam search, only applicable when temperature is zero")
-    parser.add_argument("--patience", type=float, default=None, help="optional patience value to use in beam decoding, as in https://arxiv.org/abs/2204.05424, the default (1.0) is equivalent to conventional beam search")
-    parser.add_argument("--length_penalty", type=float, default=None, help="optional token length penalty coefficient (alpha) as in https://arxiv.org/abs/1609.08144, uses simple length normalization by default")
-
-    parser.add_argument("--suppress_tokens", type=str, default="-1", help="comma-separated list of token ids to suppress during sampling; '-1' will suppress most special characters except common punctuations")
-    parser.add_argument("--initial_prompt", type=str, default=None, help="optional text to provide as a prompt for the first window.")
-    parser.add_argument("--condition_on_previous_text", type=str2bool, default=False, help="if True, provide the previous output of the model as a prompt for the next window; disabling may make the text inconsistent across windows, but the model becomes less prone to getting stuck in a failure loop")
-    parser.add_argument("--fp16", type=str2bool, default=True, help="whether to perform inference in fp16; True by default")
-
-    parser.add_argument("--temperature_increment_on_fallback", type=optional_float, default=0.2, help="temperature to increase when falling back when the decoding fails to meet either of the thresholds below")
-    parser.add_argument("--compression_ratio_threshold", type=optional_float, default=2.4, help="if the gzip compression ratio is higher than this value, treat the decoding as failed")
-    parser.add_argument("--logprob_threshold", type=optional_float, default=-1.0, help="if the average log probability is lower than this value, treat the decoding as failed")
-    parser.add_argument("--no_speech_threshold", type=optional_float, default=0.6, help="if the probability of the <|nospeech|> token is higher than this value AND the decoding has failed due to `logprob_threshold`, consider the segment as silence")
-    parser.add_argument("--threads", type=optional_int, default=0, help="number of threads used by torch for CPU inference; supercedes MKL_NUM_THREADS/OMP_NUM_THREADS")
-    parser.add_argument("--hf_token", type=str, default=None, help="Hugging Face Access Token to access PyAnnote gated models")
-    
-    args = parser.parse_args().__dict__
     model_name: str = args.pop("model")
+    batch_size: int = args.pop("batch_size")
     model_dir: str = args.pop("model_dir")
+    model_cache_only: bool = args.pop("model_cache_only")
     output_dir: str = args.pop("output_dir")
-    output_type: str = args.pop("output_type")
+    output_format: str = args.pop("output_format")
     device: str = args.pop("device")
+    device_index: int = args.pop("device_index")
+    compute_type: str = args.pop("compute_type")
+    verbose: bool = args.pop("verbose")
+
+    # model_flush: bool = args.pop("model_flush")
+    os.makedirs(output_dir, exist_ok=True)
 
     align_model: str = args.pop("align_model")
-    align_extend: float = args.pop("align_extend")
-    align_from_prev: bool = args.pop("align_from_prev")
-    interpolate_method: bool = args.pop("interpolate_method")
-    
+    interpolate_method: str = args.pop("interpolate_method")
+    no_align: bool = args.pop("no_align")
+    task: str = args.pop("task")
+    if task == "translate":
+        # translation cannot be aligned
+        no_align = True
+
+    return_char_alignments: bool = args.pop("return_char_alignments")
+
     hf_token: str = args.pop("hf_token")
-    vad_filter: bool = args.pop("vad_filter")
-    parallel_bs: int = args.pop("parallel_bs")
+    vad_method: str = args.pop("vad_method")
+    vad_onset: float = args.pop("vad_onset")
+    vad_offset: float = args.pop("vad_offset")
+
+    chunk_size: int = args.pop("chunk_size")
 
     diarize: bool = args.pop("diarize")
     min_speakers: int = args.pop("min_speakers")
     max_speakers: int = args.pop("max_speakers")
+    diarize_model_name: str = args.pop("diarize_model")
+    print_progress: bool = args.pop("print_progress")
 
-    vad_pipeline = None
-    if vad_filter:
-        if hf_token is None:
-            print("Warning, no huggingface token used, needs to be saved in environment variable, otherwise will throw error loading VAD model...")
-        from pyannote.audio import Inference
-        vad_pipeline = Inference(
-            "pyannote/segmentation",
-            pre_aggregation_hook=lambda segmentation: segmentation,
-            use_auth_token=hf_token,
-            device=torch.device(device),
-        )
+    if args["language"] is not None:
+        args["language"] = args["language"].lower()
+        if args["language"] not in LANGUAGES:
+            if args["language"] in TO_LANGUAGE_CODE:
+                args["language"] = TO_LANGUAGE_CODE[args["language"]]
+            else:
+                raise ValueError(f"Unsupported language: {args['language']}")
 
-    diarize_pipeline = None
-    if diarize:
-        if hf_token is None:
-            print("Warning, no --hf_token used, needs to be saved in environment variable, otherwise will throw error loading diarization model...")
-        from pyannote.audio import Pipeline
-        diarize_pipeline = Pipeline.from_pretrained("pyannote/speaker-diarization@2.1",
-                                    use_auth_token=hf_token)
-
-    os.makedirs(output_dir, exist_ok=True)
-
-    if model_name.endswith(".en") and args["language"] not in {"en", "English"}:
+    if model_name.endswith(".en") and args["language"] != "en":
         if args["language"] is not None:
-            warnings.warn(f'{model_name} is an English-only model but receipted "{args["language"]}"; using English instead.')
+            warnings.warn(
+                f"{model_name} is an English-only model but received '{args['language']}'; using English instead."
+            )
         args["language"] = "en"
+    align_language = (
+        args["language"] if args["language"] is not None else "en"
+    )  # default to loading english if not specified
 
     temperature = args.pop("temperature")
-    temperature_increment_on_fallback = args.pop("temperature_increment_on_fallback")
-    if temperature_increment_on_fallback is not None:
-        temperature = tuple(np.arange(temperature, 1.0 + 1e-6, temperature_increment_on_fallback))
+    if (increment := args.pop("temperature_increment_on_fallback")) is not None:
+        temperature = tuple(np.arange(temperature, 1.0 + 1e-6, increment))
     else:
         temperature = [temperature]
 
-    threads = args.pop("threads")
-    if threads > 0:
+    faster_whisper_threads = 4
+    if (threads := args.pop("threads")) > 0:
         torch.set_num_threads(threads)
+        faster_whisper_threads = threads
 
-    from . import load_model
-    model = load_model(model_name, device=device, download_root=model_dir)
+    asr_options = {
+        "beam_size": args.pop("beam_size"),
+        "patience": args.pop("patience"),
+        "length_penalty": args.pop("length_penalty"),
+        "temperatures": temperature,
+        "compression_ratio_threshold": args.pop("compression_ratio_threshold"),
+        "log_prob_threshold": args.pop("logprob_threshold"),
+        "no_speech_threshold": args.pop("no_speech_threshold"),
+        "condition_on_previous_text": False,
+        "initial_prompt": args.pop("initial_prompt"),
+        "suppress_tokens": [int(x) for x in args.pop("suppress_tokens").split(",")],
+        "suppress_numerals": args.pop("suppress_numerals"),
+    }
 
-    align_language = args["language"] if args["language"] is not None else "en" # default to loading english if not specified
-    align_model, align_metadata = load_align_model(align_language, device, model_name=align_model)
+    writer = get_writer(output_format, output_dir)
+    word_options = ["highlight_words", "max_line_count", "max_line_width"]
+    if no_align:
+        for option in word_options:
+            if args[option]:
+                parser.error(f"--{option} not possible with --no_align")
+    if args["max_line_count"] and not args["max_line_width"]:
+        warnings.warn("--max_line_count has no effect without --max_line_width")
+    writer_args = {arg: args.pop(arg) for arg in word_options}
+
+    # Part 1: VAD & ASR Loop
+    results = []
+    tmp_results = []
+    # model = load_model(model_name, device=device, download_root=model_dir)
+    model = load_model(
+        model_name,
+        device=device,
+        device_index=device_index,
+        download_root=model_dir,
+        compute_type=compute_type,
+        language=args["language"],
+        asr_options=asr_options,
+        vad_method=vad_method,
+        vad_options={
+            "chunk_size": chunk_size,
+            "vad_onset": vad_onset,
+            "vad_offset": vad_offset,
+        },
+        task=task,
+        local_files_only=model_cache_only,
+        threads=faster_whisper_threads,
+    )
 
     for audio_path in args.pop("audio"):
-        if vad_filter:
-            if parallel_bs > 1:
-                print("Performing VAD and parallel transcribing ...")
-                result = transcribe_with_vad_parallel(model, audio_path, vad_pipeline, temperature=temperature, batch_size=parallel_bs, **args)
+        audio = load_audio(audio_path)
+        # >> VAD & ASR
+        print(">>Performing transcription...")
+        result: TranscriptionResult = model.transcribe(
+            audio,
+            batch_size=batch_size,
+            chunk_size=chunk_size,
+            print_progress=print_progress,
+            verbose=verbose,
+        )
+        results.append((result, audio_path))
+
+    # Unload Whisper and VAD
+    del model
+    gc.collect()
+    torch.cuda.empty_cache()
+
+    # Part 2: Align Loop
+    if not no_align:
+        tmp_results = results
+        results = []
+        align_model, align_metadata = load_align_model(
+            align_language, device, model_name=align_model
+        )
+        for result, audio_path in tmp_results:
+            # >> Align
+            if len(tmp_results) > 1:
+                input_audio = audio_path
             else:
-                print("Performing VAD...")
-                result = transcribe_with_vad(model, audio_path, vad_pipeline, temperature=temperature, **args)
-        else:
-            print("Performing transcription...")
-            result = transcribe(model, audio_path, temperature=temperature, **args)
+                # lazily load audio from part 1
+                input_audio = audio
 
-        if result["language"] != align_metadata["language"]:
-            # load new language
-            print(f"New language found ({result['language']})! Previous was ({align_metadata['language']}), loading new alignment model for new language...")
-            align_model, align_metadata = load_align_model(result["language"], device)
+            if align_model is not None and len(result["segments"]) > 0:
+                if result.get("language", "en") != align_metadata["language"]:
+                    # load new language
+                    print(
+                        f"New language found ({result['language']})! Previous was ({align_metadata['language']}), loading new alignment model for new language..."
+                    )
+                    align_model, align_metadata = load_align_model(
+                        result["language"], device
+                    )
+                print(">>Performing alignment...")
+                result: AlignedTranscriptionResult = align(
+                    result["segments"],
+                    align_model,
+                    align_metadata,
+                    input_audio,
+                    device,
+                    interpolate_method=interpolate_method,
+                    return_char_alignments=return_char_alignments,
+                    print_progress=print_progress,
+                )
 
+            results.append((result, audio_path))
 
-        print("Performing alignment...")
-        result_aligned = align(result["segments"], align_model, align_metadata, audio_path, device,
-                                extend_duration=align_extend, start_from_previous=align_from_prev, interpolate_method=interpolate_method)
-        audio_basename = os.path.basename(audio_path)
+        # Unload align model
+        del align_model
+        gc.collect()
+        torch.cuda.empty_cache()
 
-        if diarize:
-            print("Performing diarization...")
-            diarize_segments = diarize_pipeline(audio_path, min_speakers=min_speakers, max_speakers=max_speakers)
-            diarize_df = pd.DataFrame(diarize_segments.itertracks(yield_label=True))
-            diarize_df['start'] = diarize_df[0].apply(lambda x: x.start)
-            diarize_df['end'] = diarize_df[0].apply(lambda x: x.end)
-            # assumes each utterance is single speaker (needs fix)
-            result_segments, word_segments = assign_word_speakers(diarize_df, result_aligned["segments"], fill_nearest=True)
-            result_aligned["segments"] = result_segments
-            result_aligned["word_segments"] = word_segments
-
-        # save TXT
-        if output_type in ["txt", "all"]:
-            with open(os.path.join(output_dir, audio_basename + ".txt"), "w", encoding="utf-8") as txt:
-                write_txt(result_aligned["segments"], file=txt)
-
-        # save VTT
-        if output_type in ["vtt", "all"]:
-            with open(os.path.join(output_dir, audio_basename + ".vtt"), "w", encoding="utf-8") as vtt:
-                write_vtt(result_aligned["segments"], file=vtt)
-
-        # save SRT
-        if output_type in ["srt", "all"]:
-            with open(os.path.join(output_dir, audio_basename + ".srt"), "w", encoding="utf-8") as srt:
-                write_srt(result_aligned["segments"], file=srt)
-
-        # save TSV
-        if output_type in ["tsv", "all"]:
-            with open(os.path.join(output_dir, audio_basename + ".tsv"), "w", encoding="utf-8") as srt:
-                write_tsv(result_aligned["segments"], file=srt)
-
-        # save SRT word-level
-        if output_type in ["srt-word", "all"]:
-            # save per-word SRT
-            with open(os.path.join(output_dir, audio_basename + ".word.srt"), "w", encoding="utf-8") as srt:
-                write_srt(result_aligned["word_segments"], file=srt)
-
-        # save ASS
-        if output_type in ["ass", "all"]:
-            with open(os.path.join(output_dir, audio_basename + ".ass"), "w", encoding="utf-8") as ass:
-                write_ass(result_aligned["segments"], file=ass)
-        
-        # # save ASS character-level
-        if output_type in ["ass-char"]:
-            with open(os.path.join(output_dir, audio_basename + ".char.ass"), "w", encoding="utf-8") as ass:
-                write_ass(result_aligned["segments"], file=ass, resolution="char")
-
-        # save word tsv
-        if output_type in ["pickle"]:
-            exp_fp = os.path.join(output_dir, audio_basename + ".pkl")
-            pd.DataFrame(result_aligned["segments"]).to_pickle(exp_fp)
-
-        # save word tsv
-        if output_type in ["vad"]:
-            exp_fp = os.path.join(output_dir, audio_basename + ".sad")
-            wrd_segs = pd.concat([x["word-segments"] for x in result_aligned["segments"]])[['start','end']]
-            wrd_segs.to_csv(exp_fp, sep='\t', header=None, index=False)
-if __name__ == "__main__":
-    cli()
+    # >> Diarize
+    if diarize:
+        if hf_token is None:
+            print(
+                "Warning, no --hf_token used, needs to be saved in environment variable, otherwise will throw error loading diarization model..."
+            )
+        tmp_results = results
+        print(">>Performing diarization...")
+        print(">>Using model:", diarize_model_name)
+        results = []
+        diarize_model = DiarizationPipeline(model_name=diarize_model_name, use_auth_token=hf_token, device=device)
+        for result, input_audio_path in tmp_results:
+            diarize_segments = diarize_model(
+                input_audio_path, min_speakers=min_speakers, max_speakers=max_speakers
+            )
+            result = assign_word_speakers(diarize_segments, result)
+            results.append((result, input_audio_path))
+    # >> Write
+    for result, audio_path in results:
+        result["language"] = align_language
+        writer(result, audio_path, writer_args)

whisperx/types.py

@@ -0,0 +1,69 @@
+from typing import TypedDict, Optional, List, Tuple
+
+
+class SingleWordSegment(TypedDict):
+    """
+    A single word of a speech.
+    """
+    word: str
+    start: float
+    end: float
+    score: float
+
+class SingleCharSegment(TypedDict):
+    """
+    A single char of a speech.
+    """
+    char: str
+    start: float
+    end: float
+    score: float
+
+
+class SingleSegment(TypedDict):
+    """
+    A single segment (up to multiple sentences) of a speech.
+    """
+
+    start: float
+    end: float
+    text: str
+
+
+class SegmentData(TypedDict):
+    """
+    Temporary processing data used during alignment.
+    Contains cleaned and preprocessed data for each segment.
+    """
+    clean_char: List[str]  # Cleaned characters that exist in model dictionary
+    clean_cdx: List[int]   # Original indices of cleaned characters
+    clean_wdx: List[int]   # Indices of words containing valid characters
+    sentence_spans: List[Tuple[int, int]]  # Start and end indices of sentences
+
+
+class SingleAlignedSegment(TypedDict):
+    """
+    A single segment (up to multiple sentences) of a speech with word alignment.
+    """
+
+    start: float
+    end: float
+    text: str
+    words: List[SingleWordSegment]
+    chars: Optional[List[SingleCharSegment]]
+
+
+class TranscriptionResult(TypedDict):
+    """
+    A list of segments and word segments of a speech.
+    """
+    segments: List[SingleSegment]
+    language: str
+
+
+class AlignedTranscriptionResult(TypedDict):
+    """
+    A list of segments and word segments of a speech.
+    """
+    segments: List[SingleAlignedSegment]
+    word_segments: List[SingleWordSegment]

whisperx/utils.py

@@ -1,8 +1,146 @@
+import json
 import os
+import re
+import sys
 import zlib
-from typing import Callable, TextIO, Iterator, Tuple
-import pandas as pd
-import numpy as np
+from typing import Callable, Optional, TextIO
+
+LANGUAGES = {
+    "en": "english",
+    "zh": "chinese",
+    "de": "german",
+    "es": "spanish",
+    "ru": "russian",
+    "ko": "korean",
+    "fr": "french",
+    "ja": "japanese",
+    "pt": "portuguese",
+    "tr": "turkish",
+    "pl": "polish",
+    "ca": "catalan",
+    "nl": "dutch",
+    "ar": "arabic",
+    "sv": "swedish",
+    "it": "italian",
+    "id": "indonesian",
+    "hi": "hindi",
+    "fi": "finnish",
+    "vi": "vietnamese",
+    "he": "hebrew",
+    "uk": "ukrainian",
+    "el": "greek",
+    "ms": "malay",
+    "cs": "czech",
+    "ro": "romanian",
+    "da": "danish",
+    "hu": "hungarian",
+    "ta": "tamil",
+    "no": "norwegian",
+    "th": "thai",
+    "ur": "urdu",
+    "hr": "croatian",
+    "bg": "bulgarian",
+    "lt": "lithuanian",
+    "la": "latin",
+    "mi": "maori",
+    "ml": "malayalam",
+    "cy": "welsh",
+    "sk": "slovak",
+    "te": "telugu",
+    "fa": "persian",
+    "lv": "latvian",
+    "bn": "bengali",
+    "sr": "serbian",
+    "az": "azerbaijani",
+    "sl": "slovenian",
+    "kn": "kannada",
+    "et": "estonian",
+    "mk": "macedonian",
+    "br": "breton",
+    "eu": "basque",
+    "is": "icelandic",
+    "hy": "armenian",
+    "ne": "nepali",
+    "mn": "mongolian",
+    "bs": "bosnian",
+    "kk": "kazakh",
+    "sq": "albanian",
+    "sw": "swahili",
+    "gl": "galician",
+    "mr": "marathi",
+    "pa": "punjabi",
+    "si": "sinhala",
+    "km": "khmer",
+    "sn": "shona",
+    "yo": "yoruba",
+    "so": "somali",
+    "af": "afrikaans",
+    "oc": "occitan",
+    "ka": "georgian",
+    "be": "belarusian",
+    "tg": "tajik",
+    "sd": "sindhi",
+    "gu": "gujarati",
+    "am": "amharic",
+    "yi": "yiddish",
+    "lo": "lao",
+    "uz": "uzbek",
+    "fo": "faroese",
+    "ht": "haitian creole",
+    "ps": "pashto",
+    "tk": "turkmen",
+    "nn": "nynorsk",
+    "mt": "maltese",
+    "sa": "sanskrit",
+    "lb": "luxembourgish",
+    "my": "myanmar",
+    "bo": "tibetan",
+    "tl": "tagalog",
+    "mg": "malagasy",
+    "as": "assamese",
+    "tt": "tatar",
+    "haw": "hawaiian",
+    "ln": "lingala",
+    "ha": "hausa",
+    "ba": "bashkir",
+    "jw": "javanese",
+    "su": "sundanese",
+    "yue": "cantonese",
+}
+
+# language code lookup by name, with a few language aliases
+TO_LANGUAGE_CODE = {
+    **{language: code for code, language in LANGUAGES.items()},
+    "burmese": "my",
+    "valencian": "ca",
+    "flemish": "nl",
+    "haitian": "ht",
+    "letzeburgesch": "lb",
+    "pushto": "ps",
+    "panjabi": "pa",
+    "moldavian": "ro",
+    "moldovan": "ro",
+    "sinhalese": "si",
+    "castilian": "es",
+}
+
+LANGUAGES_WITHOUT_SPACES = ["ja", "zh"]
+
+system_encoding = sys.getdefaultencoding()
+
+if system_encoding != "utf-8":
+
+    def make_safe(string):
+        # replaces any character not representable using the system default encoding with an '?',
+        # avoiding UnicodeEncodeError (https://github.com/openai/whisper/discussions/729).
+        return string.encode(system_encoding, errors="replace").decode(system_encoding)
+
+else:
+
+    def make_safe(string):
+        # utf-8 can encode any Unicode code point, so no need to do the round-trip encoding
+        return string
+
 
 def exact_div(x, y):
     assert x % y == 0
@@ -30,7 +168,9 @@ def compression_ratio(text) -> float:
     return len(text_bytes) / len(zlib.compress(text_bytes))
 
 
-def format_timestamp(seconds: float, always_include_hours: bool = False, decimal_marker: str = '.'):
+def format_timestamp(
+    seconds: float, always_include_hours: bool = False, decimal_marker: str = "."
+):
     assert seconds >= 0, "non-negative timestamp expected"
     milliseconds = round(seconds * 1000.0)
 
@@ -44,214 +184,259 @@ def format_timestamp(seconds: float, always_include_hours: bool = False, decimal
     milliseconds -= seconds * 1_000
 
     hours_marker = f"{hours:02d}:" if always_include_hours or hours > 0 else ""
-    return f"{hours_marker}{minutes:02d}:{seconds:02d}{decimal_marker}{milliseconds:03d}"
+    return (
+        f"{hours_marker}{minutes:02d}:{seconds:02d}{decimal_marker}{milliseconds:03d}"
+    )
 
 
-def write_txt(transcript: Iterator[dict], file: TextIO):
-    for segment in transcript:
-        print(segment['text'].strip(), file=file, flush=True)
+class ResultWriter:
+    extension: str
 
+    def __init__(self, output_dir: str):
+        self.output_dir = output_dir
 
-def write_vtt(transcript: Iterator[dict], file: TextIO):
-    print("WEBVTT\n", file=file)
-    for segment in transcript:
-        print(
-            f"{format_timestamp(segment['start'])} --> {format_timestamp(segment['end'])}\n"
-            f"{segment['text'].strip().replace('-->', '->')}\n",
-            file=file,
-            flush=True,
+    def __call__(self, result: dict, audio_path: str, options: dict):
+        audio_basename = os.path.basename(audio_path)
+        audio_basename = os.path.splitext(audio_basename)[0]
+        output_path = os.path.join(
+            self.output_dir, audio_basename + "." + self.extension
         )
 
-def write_tsv(transcript: Iterator[dict], file: TextIO):
-    print("start", "end", "text", sep="\t", file=file)
-    for segment in transcript:
-        print(segment['start'], file=file, end="\t")
-        print(segment['end'], file=file, end="\t")
-        print(segment['text'].strip().replace("\t", " "), file=file, flush=True)
+        with open(output_path, "w", encoding="utf-8") as f:
+            self.write_result(result, file=f, options=options)
+
+    def write_result(self, result: dict, file: TextIO, options: dict):
+        raise NotImplementedError
 
 
-def write_srt(transcript: Iterator[dict], file: TextIO):
-    """
-    Write a transcript to a file in SRT format.
+class WriteTXT(ResultWriter):
+    extension: str = "txt"
 
-    Example usage:
-        from pathlib import Path
-        from whisper.utils import write_srt
-
-        result = transcribe(model, audio_path, temperature=temperature, **args)
-
-        # save SRT
-        audio_basename = Path(audio_path).stem
-        with open(Path(output_dir) / (audio_basename + ".srt"), "w", encoding="utf-8") as srt:
-            write_srt(result["segments"], file=srt)
-    """
-    for i, segment in enumerate(transcript, start=1):
-        # write srt lines
-        print(
-            f"{i}\n"
-            f"{format_timestamp(segment['start'], always_include_hours=True, decimal_marker=',')} --> "
-            f"{format_timestamp(segment['end'], always_include_hours=True, decimal_marker=',')}\n"
-            f"{segment['text'].strip().replace('-->', '->')}\n",
-            file=file,
-            flush=True,
-        )
-
-
-def write_ass(transcript: Iterator[dict],
-            file: TextIO,
-            resolution: str = "word",
-            color: str = None, underline=True,
-            prefmt: str = None, suffmt: str = None,
-            font: str = None, font_size: int = 24,
-            strip=True, **kwargs):
-    """
-    Credit: https://github.com/jianfch/stable-ts/blob/ff79549bd01f764427879f07ecd626c46a9a430a/stable_whisper/text_output.py
-        Generate Advanced SubStation Alpha (ass) file from results to
-    display both phrase-level & word-level timestamp simultaneously by:
-     -using segment-level timestamps display phrases as usual
-     -using word-level timestamps change formats (e.g. color/underline) of the word in the displayed segment
-    Note: ass file is used in the same way as srt, vtt, etc.
-    Parameters
-    ----------
-    transcript: dict
-        results from modified model
-    file: TextIO
-        file object to write to
-    resolution: str
-        "word" or "char", timestamp resolution to highlight.
-    color: str
-        color code for a word at its corresponding timestamp
-        <bbggrr> reverse order hexadecimal RGB value (e.g. FF0000 is full intensity blue. Default: 00FF00)
-    underline: bool
-        whether to underline a word at its corresponding timestamp
-    prefmt: str
-        used to specify format for word-level timestamps (must be use with 'suffmt' and overrides 'color'&'underline')
-        appears as such in the .ass file:
-            Hi, {<prefmt>}how{<suffmt>} are you?
-        reference [Appendix A: Style override codes] in http://www.tcax.org/docs/ass-specs.htm
-    suffmt: str
-        used to specify format for word-level timestamps (must be use with 'prefmt' and overrides 'color'&'underline')
-        appears as such in the .ass file:
-            Hi, {<prefmt>}how{<suffmt>} are you?
-        reference [Appendix A: Style override codes] in http://www.tcax.org/docs/ass-specs.htm
-    font: str
-        word font (default: Arial)
-    font_size: int
-        word font size (default: 48)
-    kwargs:
-        used for format styles:
-        'Name', 'Fontname', 'Fontsize', 'PrimaryColour', 'SecondaryColour', 'OutlineColour', 'BackColour', 'Bold',
-        'Italic', 'Underline', 'StrikeOut', 'ScaleX', 'ScaleY', 'Spacing', 'Angle', 'BorderStyle', 'Outline',
-        'Shadow', 'Alignment', 'MarginL', 'MarginR', 'MarginV', 'Encoding'
-
-    """
-
-    fmt_style_dict = {'Name': 'Default', 'Fontname': 'Arial', 'Fontsize': '48', 'PrimaryColour': '&Hffffff',
-                    'SecondaryColour': '&Hffffff', 'OutlineColour': '&H0', 'BackColour': '&H0', 'Bold': '0',
-                    'Italic': '0', 'Underline': '0', 'StrikeOut': '0', 'ScaleX': '100', 'ScaleY': '100',
-                    'Spacing': '0', 'Angle': '0', 'BorderStyle': '1', 'Outline': '1', 'Shadow': '0',
-                    'Alignment': '2', 'MarginL': '10', 'MarginR': '10', 'MarginV': '10', 'Encoding': '0'}
-
-    for k, v in filter(lambda x: 'colour' in x[0].lower() and not str(x[1]).startswith('&H'), kwargs.items()):
-        kwargs[k] = f'&H{kwargs[k]}'
-
-    fmt_style_dict.update((k, v) for k, v in kwargs.items() if k in fmt_style_dict)
-
-    if font:
-        fmt_style_dict.update(Fontname=font)
-    if font_size:
-        fmt_style_dict.update(Fontsize=font_size)
-
-    fmts = f'Format: {", ".join(map(str, fmt_style_dict.keys()))}'
-
-    styles = f'Style: {",".join(map(str, fmt_style_dict.values()))}'
-
-    ass_str = f'[Script Info]\nScriptType: v4.00+\nPlayResX: 384\nPlayResY: 288\nScaledBorderAndShadow: yes\n\n' \
-            f'[V4+ Styles]\n{fmts}\n{styles}\n\n' \
-            f'[Events]\nFormat: Layer, Start, End, Style, Name, MarginL, MarginR, MarginV, Effect, Text\n\n'
-
-    if prefmt or suffmt:
-        if suffmt:
-            assert prefmt, 'prefmt must be used along with suffmt'
-        else:
-            suffmt = r'\r'
-    else:
-        if not color:
-            color = 'HFF00'
-        underline_code = r'\u1' if underline else ''
-
-        prefmt = r'{\1c&' + f'{color.upper()}&{underline_code}' + '}'
-        suffmt = r'{\r}'
-    
-    def secs_to_hhmmss(secs: Tuple[float, int]):
-        mm, ss = divmod(secs, 60)
-        hh, mm = divmod(mm, 60)
-        return f'{hh:0>1.0f}:{mm:0>2.0f}:{ss:0>2.2f}'
-
-
-    def dialogue(chars: str, start: float, end: float, idx_0: int, idx_1: int) -> str:
-        if idx_0 == -1:
-            text = chars
-        else:
-            text = f'{chars[:idx_0]}{prefmt}{chars[idx_0:idx_1]}{suffmt}{chars[idx_1:]}'
-        return f"Dialogue: 0,{secs_to_hhmmss(start)},{secs_to_hhmmss(end)}," \
-               f"Default,,0,0,0,,{text.strip() if strip else text}"
-
-    if resolution == "word":
-        resolution_key = "word-segments"
-    elif resolution == "char":
-        resolution_key = "char-segments"
-    else:
-        raise ValueError(".ass resolution should be 'word' or 'char', not ", resolution)
-    
-    ass_arr = []
-
-    for segment in transcript:
-        # if "12" in segment['text']:
-            # import pdb; pdb.set_trace()
-        if resolution_key in segment:
-            res_segs = pd.DataFrame(segment[resolution_key])
-            prev = segment['start']
-            if "speaker" in segment:
-                speaker_str = f"[{segment['speaker']}]: "
+    def write_result(self, result: dict, file: TextIO, options: dict):
+        for segment in result["segments"]:
+            speaker = segment.get("speaker")
+            text = segment["text"].strip()
+            if speaker is not None:
+                print(f"[{speaker}]: {text}", file=file, flush=True)
             else:
-                speaker_str = ""
-            for cdx, crow in res_segs.iterrows():
-                if not np.isnan(crow['start']):
-                    if resolution == "char":
-                        idx_0 = cdx
-                        idx_1 = cdx + 1
-                    elif resolution == "word":
-                        idx_0 = int(crow["segment-text-start"])
-                        idx_1 = int(crow["segment-text-end"])
-                    # fill gap
-                    if crow['start'] > prev:
-                        filler_ts = {
-                            "chars": speaker_str + segment['text'],
-                            "start": prev,
-                            "end": crow['start'],
-                            "idx_0": -1,
-                            "idx_1": -1
-                        }
+                print(text, file=file, flush=True)
 
-                        ass_arr.append(filler_ts)
-                    # highlight current word
-                    f_word_ts = {
-                        "chars": speaker_str + segment['text'],
-                        "start": crow['start'],
-                        "end": crow['end'],
-                        "idx_0": idx_0 + len(speaker_str),
-                        "idx_1": idx_1 + len(speaker_str)
-                    }
-                    ass_arr.append(f_word_ts)
-                    prev = crow['end']
 
-    ass_str += '\n'.join(map(lambda x: dialogue(**x), ass_arr))
+class SubtitlesWriter(ResultWriter):
+    always_include_hours: bool
+    decimal_marker: str
 
-    file.write(ass_str)
+    def iterate_result(self, result: dict, options: dict):
+        raw_max_line_width: Optional[int] = options["max_line_width"]
+        max_line_count: Optional[int] = options["max_line_count"]
+        highlight_words: bool = options["highlight_words"]
+        max_line_width = 1000 if raw_max_line_width is None else raw_max_line_width
+        preserve_segments = max_line_count is None or raw_max_line_width is None
+
+        if len(result["segments"]) == 0:
+            return
+
+        def iterate_subtitles():
+            line_len = 0
+            line_count = 1
+            # the next subtitle to yield (a list of word timings with whitespace)
+            subtitle: list[dict] = []
+            times: list[tuple] = []
+            last = result["segments"][0]["start"]
+            for segment in result["segments"]:
+                for i, original_timing in enumerate(segment["words"]):
+                    timing = original_timing.copy()
+                    long_pause = not preserve_segments
+                    if "start" in timing:
+                        long_pause = long_pause and timing["start"] - last > 3.0
+                    else:
+                        long_pause = False
+                    has_room = line_len + len(timing["word"]) <= max_line_width
+                    seg_break = i == 0 and len(subtitle) > 0 and preserve_segments
+                    if line_len > 0 and has_room and not long_pause and not seg_break:
+                        # line continuation
+                        line_len += len(timing["word"])
+                    else:
+                        # new line
+                        timing["word"] = timing["word"].strip()
+                        if (
+                            len(subtitle) > 0
+                            and max_line_count is not None
+                            and (long_pause or line_count >= max_line_count)
+                            or seg_break
+                        ):
+                            # subtitle break
+                            yield subtitle, times
+                            subtitle = []
+                            times = []
+                            line_count = 1
+                        elif line_len > 0:
+                            # line break
+                            line_count += 1
+                            timing["word"] = "\n" + timing["word"]
+                        line_len = len(timing["word"].strip())
+                    subtitle.append(timing)
+                    times.append((segment["start"], segment["end"], segment.get("speaker")))
+                    if "start" in timing:
+                        last = timing["start"]
+            if len(subtitle) > 0:
+                yield subtitle, times
+
+        if "words" in result["segments"][0]:
+            for subtitle, _ in iterate_subtitles():
+                sstart, ssend, speaker = _[0]
+                subtitle_start = self.format_timestamp(sstart)
+                subtitle_end = self.format_timestamp(ssend)
+                if result["language"] in LANGUAGES_WITHOUT_SPACES:
+                    subtitle_text = "".join([word["word"] for word in subtitle])
+                else:
+                    subtitle_text = " ".join([word["word"] for word in subtitle])
+                has_timing = any(["start" in word for word in subtitle])
+
+                # add [$SPEAKER_ID]: to each subtitle if speaker is available
+                prefix = ""
+                if speaker is not None:
+                    prefix = f"[{speaker}]: "
+
+                if highlight_words and has_timing:
+                    last = subtitle_start
+                    all_words = [timing["word"] for timing in subtitle]
+                    for i, this_word in enumerate(subtitle):
+                        if "start" in this_word:
+                            start = self.format_timestamp(this_word["start"])
+                            end = self.format_timestamp(this_word["end"])
+                            if last != start:
+                                yield last, start, prefix + subtitle_text
+
+                            yield start, end, prefix + " ".join(
+                                [
+                                    re.sub(r"^(\s*)(.*)$", r"\1<u>\2</u>", word)
+                                    if j == i
+                                    else word
+                                    for j, word in enumerate(all_words)
+                                ]
+                            )
+                            last = end
+                else:
+                    yield subtitle_start, subtitle_end, prefix + subtitle_text
+        else:
+            for segment in result["segments"]:
+                segment_start = self.format_timestamp(segment["start"])
+                segment_end = self.format_timestamp(segment["end"])
+                segment_text = segment["text"].strip().replace("-->", "->")
+                if "speaker" in segment:
+                    segment_text = f"[{segment['speaker']}]: {segment_text}"
+                yield segment_start, segment_end, segment_text
+
+    def format_timestamp(self, seconds: float):
+        return format_timestamp(
+            seconds=seconds,
+            always_include_hours=self.always_include_hours,
+            decimal_marker=self.decimal_marker,
+        )
+
+
+class WriteVTT(SubtitlesWriter):
+    extension: str = "vtt"
+    always_include_hours: bool = False
+    decimal_marker: str = "."
+
+    def write_result(self, result: dict, file: TextIO, options: dict):
+        print("WEBVTT\n", file=file)
+        for start, end, text in self.iterate_result(result, options):
+            print(f"{start} --> {end}\n{text}\n", file=file, flush=True)
+
+
+class WriteSRT(SubtitlesWriter):
+    extension: str = "srt"
+    always_include_hours: bool = True
+    decimal_marker: str = ","
+
+    def write_result(self, result: dict, file: TextIO, options: dict):
+        for i, (start, end, text) in enumerate(
+            self.iterate_result(result, options), start=1
+        ):
+            print(f"{i}\n{start} --> {end}\n{text}\n", file=file, flush=True)
+
+
+class WriteTSV(ResultWriter):
+    """
+    Write a transcript to a file in TSV (tab-separated values) format containing lines like:
+    <start time in integer milliseconds>\t<end time in integer milliseconds>\t<transcript text>
+
+    Using integer milliseconds as start and end times means there's no chance of interference from
+    an environment setting a language encoding that causes the decimal in a floating point number
+    to appear as a comma; also is faster and more efficient to parse & store, e.g., in C++.
+    """
+
+    extension: str = "tsv"
+
+    def write_result(self, result: dict, file: TextIO, options: dict):
+        print("start", "end", "text", sep="\t", file=file)
+        for segment in result["segments"]:
+            print(round(1000 * segment["start"]), file=file, end="\t")
+            print(round(1000 * segment["end"]), file=file, end="\t")
+            print(segment["text"].strip().replace("\t", " "), file=file, flush=True)
+
+class WriteAudacity(ResultWriter):
+    """
+    Write a transcript to a text file that audacity can import as labels.
+    The extension used is "aud" to distinguish it from the txt file produced by WriteTXT.
+    Yet this is not an audacity project but only a label file!
+    
+    Please note : Audacity uses seconds in timestamps not ms! 
+    Also there is no header expected.
+
+    If speaker is provided it is prepended to the text between double square brackets [[]].
+    """
+
+    extension: str = "aud"    
+
+    def write_result(self, result: dict, file: TextIO, options: dict):
+        ARROW = "	"
+        for segment in result["segments"]:
+            print(segment["start"], file=file, end=ARROW)
+            print(segment["end"], file=file, end=ARROW)
+            print( ( ("[[" + segment["speaker"] + "]]") if "speaker" in segment else "") + segment["text"].strip().replace("\t", " "), file=file, flush=True)
+
+            
+
+class WriteJSON(ResultWriter):
+    extension: str = "json"
+
+    def write_result(self, result: dict, file: TextIO, options: dict):
+        json.dump(result, file, ensure_ascii=False)
+
+
+def get_writer(
+    output_format: str, output_dir: str
+) -> Callable[[dict, TextIO, dict], None]:
+    writers = {
+        "txt": WriteTXT,
+        "vtt": WriteVTT,
+        "srt": WriteSRT,
+        "tsv": WriteTSV,
+        "json": WriteJSON,
+    }
+    optional_writers = {
+        "aud": WriteAudacity,
+    }
+
+    if output_format == "all":
+        all_writers = [writer(output_dir) for writer in writers.values()]
+
+        def write_all(result: dict, file: TextIO, options: dict):
+            for writer in all_writers:
+                writer(result, file, options)
+
+        return write_all
+
+    if output_format in optional_writers:
+        return optional_writers[output_format](output_dir)
+    return writers[output_format](output_dir)
 
 def interpolate_nans(x, method='nearest'):
     if x.notnull().sum() > 1:
         return x.interpolate(method=method).ffill().bfill()
     else:
-        return x.ffill().bfill()
+        return x.ffill().bfill()

whisperx/vad.py

@@ -1,185 +0,0 @@
-import pandas as pd
-import numpy as np
-from pyannote.core import Annotation, Segment, SlidingWindowFeature, Timeline
-from typing import List, Tuple, Optional
-
-class Binarize:
-    """Binarize detection scores using hysteresis thresholding
-    Parameters
-    ----------
-    onset : float, optional
-        Onset threshold. Defaults to 0.5.
-    offset : float, optional
-        Offset threshold. Defaults to `onset`.
-    min_duration_on : float, optional
-        Remove active regions shorter than that many seconds. Defaults to 0s.
-    min_duration_off : float, optional
-        Fill inactive regions shorter than that many seconds. Defaults to 0s.
-    pad_onset : float, optional
-        Extend active regions by moving their start time by that many seconds.
-        Defaults to 0s.
-    pad_offset : float, optional
-        Extend active regions by moving their end time by that many seconds.
-        Defaults to 0s.
-    max_duration: float
-        The maximum length of an active segment, divides segment at timestamp with lowest score.
-    Reference
-    ---------
-    Gregory Gelly and Jean-Luc Gauvain. "Minimum Word Error Training of
-    RNN-based Voice Activity Detection", InterSpeech 2015.
-
-    Pyannote-audio
-    """
-
-    def __init__(
-        self,
-        onset: float = 0.5,
-        offset: Optional[float] = None,
-        min_duration_on: float = 0.0,
-        min_duration_off: float = 0.0,
-        pad_onset: float = 0.0,
-        pad_offset: float = 0.0,
-        max_duration: float = float('inf')
-    ):
-
-        super().__init__()
-
-        self.onset = onset
-        self.offset = offset or onset
-
-        self.pad_onset = pad_onset
-        self.pad_offset = pad_offset
-
-        self.min_duration_on = min_duration_on
-        self.min_duration_off = min_duration_off
-
-        self.max_duration = max_duration
-
-    def __call__(self, scores: SlidingWindowFeature) -> Annotation:
-        """Binarize detection scores
-        Parameters
-        ----------
-        scores : SlidingWindowFeature
-            Detection scores.
-        Returns
-        -------
-        active : Annotation
-            Binarized scores.
-        """
-
-        num_frames, num_classes = scores.data.shape
-        frames = scores.sliding_window
-        timestamps = [frames[i].middle for i in range(num_frames)]
-
-        # annotation meant to store 'active' regions
-        active = Annotation()
-        for k, k_scores in enumerate(scores.data.T):
-
-            label = k if scores.labels is None else scores.labels[k]
-
-            # initial state
-            start = timestamps[0]
-            is_active = k_scores[0] > self.onset
-            curr_scores = [k_scores[0]]
-            curr_timestamps = [start]
-            for t, y in zip(timestamps[1:], k_scores[1:]):
-                # currently active
-                if is_active: 
-                    curr_duration = t - start
-                    if curr_duration > self.max_duration:
-                        # if curr_duration > 15:
-                            # import pdb; pdb.set_trace()
-                        search_after = len(curr_scores) // 2
-                        # divide segment
-                        min_score_div_idx = search_after + np.argmin(curr_scores[search_after:])
-                        min_score_t = curr_timestamps[min_score_div_idx]
-                        region = Segment(start - self.pad_onset, min_score_t + self.pad_offset)
-                        active[region, k] = label
-                        start = curr_timestamps[min_score_div_idx]
-                        curr_scores = curr_scores[min_score_div_idx+1:]
-                        curr_timestamps = curr_timestamps[min_score_div_idx+1:]
-                    # switching from active to inactive
-                    elif y < self.offset:
-                        region = Segment(start - self.pad_onset, t + self.pad_offset)
-                        active[region, k] = label
-                        start = t
-                        is_active = False
-                        curr_scores = []
-                        curr_timestamps = []
-                # currently inactive
-                else:
-                    # switching from inactive to active
-                    if y > self.onset:
-                        start = t
-                        is_active = True
-                curr_scores.append(y)
-                curr_timestamps.append(t)
-
-            # if active at the end, add final region
-            if is_active:
-                region = Segment(start - self.pad_onset, t + self.pad_offset)
-                active[region, k] = label
-
-        # because of padding, some active regions might be overlapping: merge them.
-        # also: fill same speaker gaps shorter than min_duration_off
-        if self.pad_offset > 0.0 or self.pad_onset > 0.0 or self.min_duration_off > 0.0:
-            if self.max_duration < float("inf"):
-                raise NotImplementedError(f"This would break current max_duration param")
-            active = active.support(collar=self.min_duration_off)
-
-        # remove tracks shorter than min_duration_on
-        if self.min_duration_on > 0:
-            for segment, track in list(active.itertracks()):
-                if segment.duration < self.min_duration_on:
-                    del active[segment, track]
-
-        return active
-
-
-def merge_vad(vad_arr, pad_onset=0.0, pad_offset=0.0, min_duration_off=0.0, min_duration_on=0.0):
-
-    active = Annotation()
-    for k, vad_t in enumerate(vad_arr):
-        region = Segment(vad_t[0] - pad_onset, vad_t[1] + pad_offset)
-        active[region, k] = 1
-
-
-    if pad_offset > 0.0 or pad_onset > 0.0 or min_duration_off > 0.0:
-        active = active.support(collar=min_duration_off)
-    
-    # remove tracks shorter than min_duration_on
-    if min_duration_on > 0:
-        for segment, track in list(active.itertracks()):
-            if segment.duration < min_duration_on:
-                    del active[segment, track]
-    
-    active = active.for_json()
-    active_segs = pd.DataFrame([x['segment'] for x in active['content']])
-    return active_segs
-
-
-if __name__ == "__main__":
-    # from pyannote.audio import Inference
-    # hook = lambda segmentation: segmentation
-    # inference = Inference("pyannote/segmentation", pre_aggregation_hook=hook)
-    # audio = "/tmp/11962.wav" 
-    # scores = inference(audio)
-    # binarize = Binarize(max_duration=15)
-    # anno = binarize(scores)
-    # res = []
-    # for ann in anno.get_timeline():
-    #     res.append((ann.start, ann.end))
-
-    # res = pd.DataFrame(res)
-    # res[2] = res[1] - res[0]
-    import pandas as pd
-    input_fp = "tt298650_sync.wav"
-    df = pd.read_csv(f"/work/maxbain/tmp/{input_fp}.sad", sep=" ", header=None)
-    print(len(df))
-    N = 0.15
-    g = df[0].sub(df[1].shift())
-    input_base = input_fp.split('.')[0]
-    df = df.groupby(g.gt(N).cumsum()).agg({0:'min', 1:'max'})
-    df.to_csv(f"/work/maxbain/tmp/{input_base}.lab", header=None, index=False, sep=" ")
-    print(df)
-    import pdb; pdb.set_trace()

whisperx/vads/__init__.py

@@ -0,0 +1,3 @@
+from whisperx.vads.pyannote import Pyannote as Pyannote
+from whisperx.vads.silero import Silero as Silero
+from whisperx.vads.vad import Vad as Vad

whisperx/vads/pyannote.py

@@ -0,0 +1,263 @@
+import os
+from typing import Callable, Text, Union
+from typing import Optional
+
+import numpy as np
+import torch
+from pyannote.audio import Model
+from pyannote.audio.core.io import AudioFile
+from pyannote.audio.pipelines import VoiceActivityDetection
+from pyannote.audio.pipelines.utils import PipelineModel
+from pyannote.core import Annotation, SlidingWindowFeature
+from pyannote.core import Segment
+
+from whisperx.diarize import Segment as SegmentX
+from whisperx.vads.vad import Vad
+
+
+def load_vad_model(device, vad_onset=0.500, vad_offset=0.363, use_auth_token=None, model_fp=None):
+    model_dir = torch.hub._get_torch_home()
+
+    main_dir = os.path.dirname(os.path.dirname(os.path.abspath(__file__)))
+
+    os.makedirs(model_dir, exist_ok = True)
+    if model_fp is None:
+        # Dynamically resolve the path to the model file
+        model_fp = os.path.join(main_dir, "assets", "pytorch_model.bin")
+        model_fp = os.path.abspath(model_fp)  # Ensure the path is absolute
+    else:
+        model_fp = os.path.abspath(model_fp)  # Ensure any provided path is absolute
+
+    # Check if the resolved model file exists
+    if not os.path.exists(model_fp):
+        raise FileNotFoundError(f"Model file not found at {model_fp}")
+
+    if os.path.exists(model_fp) and not os.path.isfile(model_fp):
+        raise RuntimeError(f"{model_fp} exists and is not a regular file")
+
+    model_bytes = open(model_fp, "rb").read()
+
+    vad_model = Model.from_pretrained(model_fp, use_auth_token=use_auth_token)
+    hyperparameters = {"onset": vad_onset,
+                    "offset": vad_offset,
+                    "min_duration_on": 0.1,
+                    "min_duration_off": 0.1}
+    vad_pipeline = VoiceActivitySegmentation(segmentation=vad_model, device=torch.device(device))
+    vad_pipeline.instantiate(hyperparameters)
+
+    return vad_pipeline
+
+class Binarize:
+    """Binarize detection scores using hysteresis thresholding, with min-cut operation
+    to ensure not segments are longer than max_duration.
+
+    Parameters
+    ----------
+    onset : float, optional
+        Onset threshold. Defaults to 0.5.
+    offset : float, optional
+        Offset threshold. Defaults to `onset`.
+    min_duration_on : float, optional
+        Remove active regions shorter than that many seconds. Defaults to 0s.
+    min_duration_off : float, optional
+        Fill inactive regions shorter than that many seconds. Defaults to 0s.
+    pad_onset : float, optional
+        Extend active regions by moving their start time by that many seconds.
+        Defaults to 0s.
+    pad_offset : float, optional
+        Extend active regions by moving their end time by that many seconds.
+        Defaults to 0s.
+    max_duration: float
+        The maximum length of an active segment, divides segment at timestamp with lowest score.
+    Reference
+    ---------
+    Gregory Gelly and Jean-Luc Gauvain. "Minimum Word Error Training of
+    RNN-based Voice Activity Detection", InterSpeech 2015.
+
+    Modified by Max Bain to include WhisperX's min-cut operation
+    https://arxiv.org/abs/2303.00747
+
+    Pyannote-audio
+    """
+
+    def __init__(
+            self,
+            onset: float = 0.5,
+            offset: Optional[float] = None,
+            min_duration_on: float = 0.0,
+            min_duration_off: float = 0.0,
+            pad_onset: float = 0.0,
+            pad_offset: float = 0.0,
+            max_duration: float = float('inf')
+    ):
+
+        super().__init__()
+
+        self.onset = onset
+        self.offset = offset or onset
+
+        self.pad_onset = pad_onset
+        self.pad_offset = pad_offset
+
+        self.min_duration_on = min_duration_on
+        self.min_duration_off = min_duration_off
+
+        self.max_duration = max_duration
+
+    def __call__(self, scores: SlidingWindowFeature) -> Annotation:
+        """Binarize detection scores
+        Parameters
+        ----------
+        scores : SlidingWindowFeature
+            Detection scores.
+        Returns
+        -------
+        active : Annotation
+            Binarized scores.
+        """
+
+        num_frames, num_classes = scores.data.shape
+        frames = scores.sliding_window
+        timestamps = [frames[i].middle for i in range(num_frames)]
+
+        # annotation meant to store 'active' regions
+        active = Annotation()
+        for k, k_scores in enumerate(scores.data.T):
+
+            label = k if scores.labels is None else scores.labels[k]
+
+            # initial state
+            start = timestamps[0]
+            is_active = k_scores[0] > self.onset
+            curr_scores = [k_scores[0]]
+            curr_timestamps = [start]
+            t = start
+            for t, y in zip(timestamps[1:], k_scores[1:]):
+                # currently active
+                if is_active:
+                    curr_duration = t - start
+                    if curr_duration > self.max_duration:
+                        search_after = len(curr_scores) // 2
+                        # divide segment
+                        min_score_div_idx = search_after + np.argmin(curr_scores[search_after:])
+                        min_score_t = curr_timestamps[min_score_div_idx]
+                        region = Segment(start - self.pad_onset, min_score_t + self.pad_offset)
+                        active[region, k] = label
+                        start = curr_timestamps[min_score_div_idx]
+                        curr_scores = curr_scores[min_score_div_idx + 1:]
+                        curr_timestamps = curr_timestamps[min_score_div_idx + 1:]
+                    # switching from active to inactive
+                    elif y < self.offset:
+                        region = Segment(start - self.pad_onset, t + self.pad_offset)
+                        active[region, k] = label
+                        start = t
+                        is_active = False
+                        curr_scores = []
+                        curr_timestamps = []
+                    curr_scores.append(y)
+                    curr_timestamps.append(t)
+                # currently inactive
+                else:
+                    # switching from inactive to active
+                    if y > self.onset:
+                        start = t
+                        is_active = True
+
+            # if active at the end, add final region
+            if is_active:
+                region = Segment(start - self.pad_onset, t + self.pad_offset)
+                active[region, k] = label
+
+        # because of padding, some active regions might be overlapping: merge them.
+        # also: fill same speaker gaps shorter than min_duration_off
+        if self.pad_offset > 0.0 or self.pad_onset > 0.0 or self.min_duration_off > 0.0:
+            if self.max_duration < float("inf"):
+                raise NotImplementedError(f"This would break current max_duration param")
+            active = active.support(collar=self.min_duration_off)
+
+        # remove tracks shorter than min_duration_on
+        if self.min_duration_on > 0:
+            for segment, track in list(active.itertracks()):
+                if segment.duration < self.min_duration_on:
+                    del active[segment, track]
+
+        return active
+
+
+class VoiceActivitySegmentation(VoiceActivityDetection):
+    def __init__(
+            self,
+            segmentation: PipelineModel = "pyannote/segmentation",
+            fscore: bool = False,
+            use_auth_token: Union[Text, None] = None,
+            **inference_kwargs,
+    ):
+
+        super().__init__(segmentation=segmentation, fscore=fscore, use_auth_token=use_auth_token, **inference_kwargs)
+
+    def apply(self, file: AudioFile, hook: Optional[Callable] = None) -> Annotation:
+        """Apply voice activity detection
+
+        Parameters
+        ----------
+        file : AudioFile
+            Processed file.
+        hook : callable, optional
+            Hook called after each major step of the pipeline with the following
+            signature: hook("step_name", step_artefact, file=file)
+
+        Returns
+        -------
+        speech : Annotation
+            Speech regions.
+        """
+
+        # setup hook (e.g. for debugging purposes)
+        hook = self.setup_hook(file, hook=hook)
+
+        # apply segmentation model (only if needed)
+        # output shape is (num_chunks, num_frames, 1)
+        if self.training:
+            if self.CACHED_SEGMENTATION in file:
+                segmentations = file[self.CACHED_SEGMENTATION]
+            else:
+                segmentations = self._segmentation(file)
+                file[self.CACHED_SEGMENTATION] = segmentations
+        else:
+            segmentations: SlidingWindowFeature = self._segmentation(file)
+
+        return segmentations
+
+
+class Pyannote(Vad):
+
+    def __init__(self, device, use_auth_token=None, model_fp=None, **kwargs):
+        print(">>Performing voice activity detection using Pyannote...")
+        super().__init__(kwargs['vad_onset'])
+        self.vad_pipeline = load_vad_model(device, use_auth_token=use_auth_token, model_fp=model_fp)
+
+    def __call__(self, audio: AudioFile, **kwargs):
+        return self.vad_pipeline(audio)
+
+    @staticmethod
+    def preprocess_audio(audio):
+        return torch.from_numpy(audio).unsqueeze(0)
+
+    @staticmethod
+    def merge_chunks(segments,
+                     chunk_size,
+                     onset: float = 0.5,
+                     offset: Optional[float] = None,
+                     ):
+        assert chunk_size > 0
+        binarize = Binarize(max_duration=chunk_size, onset=onset, offset=offset)
+        segments = binarize(segments)
+        segments_list = []
+        for speech_turn in segments.get_timeline():
+            segments_list.append(SegmentX(speech_turn.start, speech_turn.end, "UNKNOWN"))
+
+        if len(segments_list) == 0:
+            print("No active speech found in audio")
+            return []
+        assert segments_list, "segments_list is empty."
+        return Vad.merge_chunks(segments_list, chunk_size, onset, offset)

whisperx/vads/silero.py

@@ -0,0 +1,66 @@
+from io import IOBase
+from pathlib import Path
+from typing import Mapping, Text
+from typing import Optional
+from typing import Union
+
+import torch
+
+from whisperx.diarize import Segment as SegmentX
+from whisperx.vads.vad import Vad
+
+AudioFile = Union[Text, Path, IOBase, Mapping]
+
+
+class Silero(Vad):
+    # check again default values
+    def __init__(self, **kwargs):
+        print(">>Performing voice activity detection using Silero...")
+        super().__init__(kwargs['vad_onset'])
+
+        self.vad_onset = kwargs['vad_onset']
+        self.chunk_size = kwargs['chunk_size']
+        self.vad_pipeline, vad_utils = torch.hub.load(repo_or_dir='snakers4/silero-vad',
+                                                      model='silero_vad',
+                                                      force_reload=False,
+                                                      onnx=False,
+                                                      trust_repo=True)
+        (self.get_speech_timestamps, _, self.read_audio, _, _) = vad_utils
+
+    def __call__(self, audio: AudioFile, **kwargs):
+        """use silero to get segments of speech"""
+        # Only accept 16000 Hz for now.
+        # Note: Silero models support both 8000 and 16000 Hz. Although other values are not directly supported,
+        # multiples of 16000 (e.g. 32000 or 48000) are cast to 16000 inside of the JIT model!
+        sample_rate = audio["sample_rate"]
+        if sample_rate != 16000:
+            raise ValueError("Only 16000Hz sample rate is allowed")
+
+        timestamps = self.get_speech_timestamps(audio["waveform"],
+                                                model=self.vad_pipeline,
+                                                sampling_rate=sample_rate,
+                                                max_speech_duration_s=self.chunk_size,
+                                                threshold=self.vad_onset
+                                                # min_silence_duration_ms = self.min_duration_off/1000
+                                                # min_speech_duration_ms = self.min_duration_on/1000
+                                                # ...
+                                                # See silero documentation for full option list
+                                                )
+        return [SegmentX(i['start'] / sample_rate, i['end'] / sample_rate, "UNKNOWN") for i in timestamps]
+
+    @staticmethod
+    def preprocess_audio(audio):
+        return audio
+
+    @staticmethod
+    def merge_chunks(segments_list,
+                     chunk_size,
+                     onset: float = 0.5,
+                     offset: Optional[float] = None,
+                     ):
+        assert chunk_size > 0
+        if len(segments_list) == 0:
+            print("No active speech found in audio")
+            return []
+        assert segments_list, "segments_list is empty."
+        return Vad.merge_chunks(segments_list, chunk_size, onset, offset)

whisperx/vads/vad.py

@@ -0,0 +1,74 @@
+from typing import Optional
+
+import pandas as pd
+from pyannote.core import Annotation, Segment
+
+
+class Vad:
+    def __init__(self, vad_onset):
+        if not (0 < vad_onset < 1):
+            raise ValueError(
+                "vad_onset is a decimal value between 0 and 1."
+            )
+
+    @staticmethod
+    def preprocess_audio(audio):
+        pass
+
+    # keep merge_chunks as static so it can be also used by manually assigned vad_model (see 'load_model')
+    @staticmethod
+    def merge_chunks(segments,
+                     chunk_size,
+                     onset: float,
+                     offset: Optional[float]):
+        """
+         Merge operation described in paper
+         """
+        curr_end = 0
+        merged_segments = []
+        seg_idxs: list[tuple]= []
+        speaker_idxs: list[Optional[str]] = []
+
+        curr_start = segments[0].start
+        for seg in segments:
+            if seg.end - curr_start > chunk_size and curr_end - curr_start > 0:
+                merged_segments.append({
+                    "start": curr_start,
+                    "end": curr_end,
+                    "segments": seg_idxs,
+                })
+                curr_start = seg.start
+                seg_idxs = []
+                speaker_idxs = []
+            curr_end = seg.end
+            seg_idxs.append((seg.start, seg.end))
+            speaker_idxs.append(seg.speaker)
+        # add final
+        merged_segments.append({
+            "start": curr_start,
+            "end": curr_end,
+            "segments": seg_idxs,
+        })
+
+        return merged_segments
+
+    # Unused function
+    @staticmethod
+    def merge_vad(vad_arr, pad_onset=0.0, pad_offset=0.0, min_duration_off=0.0, min_duration_on=0.0):
+        active = Annotation()
+        for k, vad_t in enumerate(vad_arr):
+            region = Segment(vad_t[0] - pad_onset, vad_t[1] + pad_offset)
+            active[region, k] = 1
+
+        if pad_offset > 0.0 or pad_onset > 0.0 or min_duration_off > 0.0:
+            active = active.support(collar=min_duration_off)
+
+        # remove tracks shorter than min_duration_on
+        if min_duration_on > 0:
+            for segment, track in list(active.itertracks()):
+                if segment.duration < min_duration_on:
+                    del active[segment, track]
+
+        active = active.for_json()
+        active_segs = pd.DataFrame([x['segment'] for x in active['content']])
+        return active_segs