Merge pull request #96 from smly/fix-batch-processing

FIX: Assertion error in batch processing

.github/FUNDING.yml

@@ -0,0 +1 @@
+custom: https://www.buymeacoffee.com/maxhbain

EXAMPLES.md

@@ -2,7 +2,7 @@
 
 ## Other Languages
 
-For non-english ASR, it is best to use the `large` whisper model. Alignment models are automatically picked by the chosen language from the default [lists](https://github.com/m-bain/whisperX/blob/e909f2f766b23b2000f2d95df41f9b844ac53e49/whisperx/transcribe.py#L22).
+For non-english ASR, it is best to use the `large` whisper model. Alignment models are automatically picked by the chosen language from the default [lists](https://github.com/m-bain/whisperX/blob/main/whisperx/alignment.py#L18).
 
 Currently support default models tested for {en, fr, de, es, it, ja, zh, nl}
 

README.md

@@ -27,7 +27,6 @@
   <a href="EXAMPLES.md">More examples</a>
 </p>
 
-<h6 align="center">Made by Max Bain • :globe_with_meridians: <a href="https://www.maxbain.com">https://www.maxbain.com</a></h6>
 
 <img width="1216" align="center" alt="whisperx-arch" src="https://user-images.githubusercontent.com/36994049/211200186-8b779e26-0bfd-4127-aee2-5a9238b95e1f.png">
 
@@ -50,9 +49,10 @@ This repository refines the timestamps of openAI's Whisper model via forced alig
 
 <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 `--diarization`)
+- Diarization (still in beta, add `--diarize`)
 
 
 <h2 align="left" id="setup">Setup ⚙️</h2>
@@ -71,9 +71,13 @@ $ cd whisperX
 $ pip install -e .
 ```
 
-
 You may also need to install ffmpeg, rust etc. Follow openAI instructions here https://github.com/openai/whisper#setup.
 
+
+### 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)
+
+
 <h2 align="left" id="example">Usage 💬 (command line)</h2>
 
 ### English
@@ -85,7 +89,7 @@ Run whisper on example segment (using default params)
 
 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.en --vad_filter --align_model WAV2VEC2_ASR_LARGE_LV60K_960H
+    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:
 
@@ -149,8 +153,9 @@ In addition to forced alignment, the following two modifications have been made
 
 - 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.
-- 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)
+- 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)
-- Hacked this up quite quickly, there might be some errors, please raise an issue if you encounter any.
+- Overlapping speech is not handled particularly well by whisper nor whisperx
+- Diariazation is far from perfect.
 
 
 <h2 align="left" id="contribute">Contribute 🧑‍🏫</h2>
@@ -161,38 +166,46 @@ The next major upgrade we are working on is whisper with speaker diarization, so
 
 <h2 align="left" id="coming-soon">Coming Soon 🗓</h2>
 
-[x] ~~Multilingual init~~ done
+* [x] Multilingual init
 
-[x] ~~Subtitle .ass output~~ done
+* [x] Subtitle .ass output
 
-[x] ~~Automatic align model selection based on language detection~~ done
+* [x] Automatic align model selection based on language detection
 
-[x] ~~Python usage~~ done
+* [x] Python usage
 
-[x] ~~Character level timestamps~~
+* [x] Character level timestamps
 
-[x] ~~Incorporating  speaker diarization~~
+* [x] Incorporating  speaker diarization
 
-[ ] Improve diarization (word level)
+* [x] Inference speedup with batch processing
 
-[ ] Inference speedup with batch processing
+* [ ] Improve diarization (word level). *Harder than first thought...*
 
-<h2 align="left" id="contact">Contact 📇</h2>
 
-Contact maxbain[at]robots[dot]ox[dot]ac[dot]uk for business things.
+<h2 align="left" id="contact">Contact/Support 📇</h2>
+
+Contact maxbain[at]robots[dot]ox[dot]ac[dot]uk 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>
 
-Of course, this is mostly just a modification to [openAI's whisper](https://github.com/openai/whisper).
+This work, and my PhD, is supported by the [VGG (Visual Geometry Group)](https://www.robots.ox.ac.uk/~vgg/) and University of Oxford.
-As well as accreditation to this [PyTorch tutorial on forced alignment](https://pytorch.org/tutorials/intermediate/forced_alignment_with_torchaudio_tutorial.html)
+
+
+
+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)
 
 
 <h2 align="left" id="cite">Citation</h2>
-If you use this in your research, just cite the repo,
+If you use this in your research, for now just cite the repo,
 
 ```bibtex
 @misc{bain2022whisperx,
-  author = {Bain, Max},
+  author = {Bain, Max and Han, Tengda},
   title = {WhisperX},
   year = {2022},
   publisher = {GitHub},

whisperx/__init__.py

@@ -11,8 +11,8 @@ 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, load_align_model, align, transcribe_with_vad
+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",

whisperx/alignment.py

@@ -1,9 +1,443 @@
+""""
+Forced Alignment with Whisper
+C. Max Bain
+"""
+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
+
+
+LANGUAGES_WITHOUT_SPACES = ["ja", "zh"]
+
+DEFAULT_ALIGN_MODELS_TORCH = {
+    "en": "WAV2VEC2_ASR_BASE_960H",
+    "fr": "VOXPOPULI_ASR_BASE_10K_FR",
+    "de": "VOXPOPULI_ASR_BASE_10K_DE",
+    "es": "VOXPOPULI_ASR_BASE_10K_ES",
+    "it": "VOXPOPULI_ASR_BASE_10K_IT",
+}
+
+DEFAULT_ALIGN_MODELS_HF = {
+    "ja": "jonatasgrosman/wav2vec2-large-xlsr-53-japanese",
+    "zh": "jonatasgrosman/wav2vec2-large-xlsr-53-chinese-zh-cn",
+    "nl": "jonatasgrosman/wav2vec2-large-xlsr-53-dutch",
+    "uk": "Yehor/wav2vec2-xls-r-300m-uk-with-small-lm",
+    "pt": "jonatasgrosman/wav2vec2-large-xlsr-53-portuguese",
+    "ar": "jonatasgrosman/wav2vec2-large-xlsr-53-arabic",
+    "ru": "jonatasgrosman/wav2vec2-large-xlsr-53-russian",
+    "pl": "jonatasgrosman/wav2vec2-large-xlsr-53-polish",
+    "hu": "jonatasgrosman/wav2vec2-large-xlsr-53-hungarian",
+    "fi": "jonatasgrosman/wav2vec2-large-xlsr-53-finnish",
+    "fa": "jonatasgrosman/wav2vec2-large-xlsr-53-persian",
+    "el": "jonatasgrosman/wav2vec2-large-xlsr-53-greek",
+    "tr": "mpoyraz/wav2vec2-xls-r-300m-cv7-turkish",
+}
+
+
+def load_align_model(language_code, device, model_name=None):
+    if model_name is None:
+        # use default model
+        if language_code in DEFAULT_ALIGN_MODELS_TORCH:
+            model_name = DEFAULT_ALIGN_MODELS_TORCH[language_code]
+        elif language_code in DEFAULT_ALIGN_MODELS_HF:
+            model_name = DEFAULT_ALIGN_MODELS_HF[language_code]
+        else:
+            print(f"There is no default alignment model set for this language ({language_code}).\
+                Please find a wav2vec2.0 model finetuned on this language in https://huggingface.co/models, then pass the model name in --align_model [MODEL_NAME]")
+            raise ValueError(f"No default align-model for language: {language_code}")
+
+    if model_name in torchaudio.pipelines.__all__:
+        pipeline_type = "torchaudio"
+        bundle = torchaudio.pipelines.__dict__[model_name]
+        align_model = bundle.get_model().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)
+        except Exception as e:
+            print(e)
+            print(f"Error loading model from huggingface, check https://huggingface.co/models for finetuned wav2vec2.0 models")
+            raise ValueError(f'The chosen align_model "{model_name}" could not be found in huggingface (https://huggingface.co/models) or torchaudio (https://pytorch.org/audio/stable/pipelines.html#id14)')
+        pipeline_type = "huggingface"
+        align_model = align_model.to(device)
+        labels = processor.tokenizer.get_vocab()
+        align_dictionary = {char.lower(): code for char,code in processor.tokenizer.get_vocab().items()}
+
+    align_metadata = {"language": language_code, "dictionary": align_dictionary, "type": pipeline_type}
+
+    return align_model, align_metadata
+
+
+def align(
+    transcript: Iterator[dict],
+    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",
+):
+    """
+    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": [],
+    }
+
+    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)
+            
+
+            # 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
+
+            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
+
+                
+                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"]
+            }
+        )
+
+    
+    return {"segments": aligned_segments, "word_segments": aligned_segments_word}
+
+
 """
 source: https://pytorch.org/tutorials/intermediate/forced_alignment_with_torchaudio_tutorial.html
 """
-import torch
-from dataclasses import dataclass
-
 def get_trellis(emission, tokens, blank_id=0):
     num_frame = emission.size(0)
     num_tokens = len(tokens)

whisperx/diarize.py

@@ -0,0 +1,56 @@
+import numpy as np
+import pandas as pd
+
+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]
+
+    # 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
+
+    return result_segments, word_seg
+
+class Segment:
+    def __init__(self, start, end, speaker=None):
+        self.start = start
+        self.end = end
+        self.speaker = speaker

whisperx/transcribe.py

@@ -5,36 +5,19 @@ from typing import List, Optional, Tuple, Union, Iterator, TYPE_CHECKING
 
 import numpy as np
 import torch
-import torchaudio
-from transformers import AutoProcessor, Wav2Vec2ForCTC
 import tqdm
 from .audio import SAMPLE_RATE, N_FRAMES, HOP_LENGTH, CHUNK_LENGTH, pad_or_trim, log_mel_spectrogram, load_audio
-from .alignment import get_trellis, backtrack, merge_repeats, merge_words
+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
 
-LANGUAGES_WITHOUT_SPACES = ["ja", "zh"]
-
-DEFAULT_ALIGN_MODELS_TORCH = {
-    "en": "WAV2VEC2_ASR_BASE_960H",
-    "fr": "VOXPOPULI_ASR_BASE_10K_FR",
-    "de": "VOXPOPULI_ASR_BASE_10K_DE",
-    "es": "VOXPOPULI_ASR_BASE_10K_ES",
-    "it": "VOXPOPULI_ASR_BASE_10K_IT",
-}
-
-DEFAULT_ALIGN_MODELS_HF = {
-    "ja": "jonatasgrosman/wav2vec2-large-xlsr-53-japanese",
-    "zh": "jonatasgrosman/wav2vec2-large-xlsr-53-chinese-zh-cn",
-    "nl": "jonatasgrosman/wav2vec2-large-xlsr-53-dutch",
-    "uk": "Yehor/wav2vec2-xls-r-300m-uk-with-small-lm",
-    "pt": "jonatasgrosman/wav2vec2-large-xlsr-53-portuguese",
-}
 
 
 def transcribe(
@@ -273,362 +256,29 @@ def transcribe(
     return dict(text=tokenizer.decode(all_tokens[len(initial_prompt):]), segments=all_segments, language=language)
 
 
-def align(
-    transcript: Iterator[dict],
-    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",
-):
-    """
-    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
-
-    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
-    for segment in transcript:
-        aligned_subsegments = []
-        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("Failed to align segment: 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]
-
-            # 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("Failed to align segment: backtrack failed, resorting to original...")
-                break
-            char_segments = merge_repeats(path, transcription_cleaned)
-            # word_segments = merge_words(char_segments)
-            
-
-            # 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
-
-            char_level = {
-                "start": [],
-                "end": [],
-                "score": [],
-                "word-index": [],
-            }
-
-            word_level = {
-                "start": [],
-                "end": [],
-                "score": [],
-                "segment-text-start": [],
-                "segment-text-end": []
-            }
-
-            wdx = 0
-            seg_start_actual, seg_end_actual = None, None
-            duration = t2 - t1
-            ratio = duration * waveform_segment.size(0) / (trellis.size(0) - 1)
-            cdx_prev = 0
-            for cdx, char in enumerate(transcription + " "):
-                is_last = False
-                if cdx == len(transcription):
-                    break
-                elif cdx+1 == len(transcription):
-                    is_last = True
-
-                
-                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_level["start"].append(start)
-                char_level["end"].append(end)
-                char_level["score"].append(score)
-                char_level["word-index"].append(wdx)
-
-                # 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
-                    word_level["start"].append(None)
-                    word_level["end"].append(None)
-                    word_level["score"].append(None)
-                    word_level["segment-text-start"].append(cdx_prev-seg_lens_cumsum[sub_seg_idx])
-                    word_level["segment-text-end"].append(cdx+1-seg_lens_cumsum[sub_seg_idx])
-                    cdx_prev = cdx+2
-
-                if is_last or cdx == seg_lens_cumsum[sub_seg_idx+1] - 1:
-                    if model_lang not in LANGUAGES_WITHOUT_SPACES:
-                        char_level = pd.DataFrame(char_level)
-                        word_level = pd.DataFrame(word_level)
-                        
-                        not_space = pd.Series(list(segment["seg-text"][sub_seg_idx])) != " "
-                        word_level["start"] = char_level[not_space].groupby("word-index")["start"].min() # take min of all chars in a word ignoring space
-                        word_level["end"] = char_level[not_space].groupby("word-index")["end"].max() # take max of all chars in a word
-
-                        # fill missing
-                        if interpolate_method != "ignore":
-                            word_level["start"] = interpolate_nans(word_level["start"], method=interpolate_method) 
-                            word_level["end"] = interpolate_nans(word_level["end"], method=interpolate_method)
-                        word_level["start"] = word_level["start"].values.tolist()
-                        word_level["end"] = word_level["end"].values.tolist()
-                        word_level["score"] = char_level.groupby("word-index")["score"].mean() # take mean of all scores
-
-                        char_level = char_level.replace({np.nan:None}).to_dict("list")
-                        word_level = pd.DataFrame(word_level).replace({np.nan:None}).to_dict("list")
-                    else:
-                        word_level = None
-
-                    aligned_subsegments.append(
-                        {
-                            "text": segment["seg-text"][sub_seg_idx],
-                            "start": seg_start_actual,
-                            "end": seg_end_actual,
-                            "char-segments": char_level,
-                            "word-segments": word_level
-                        }
-                    )
-                    if "language" in segment:
-                        aligned_subsegments[-1]["language"] = segment["language"]
-
-                    char_level = {
-                        "start": [],
-                        "end": [],
-                        "score": [],
-                        "word-index": [],
-                    }
-                    word_level = {
-                        "start": [],
-                        "end": [],
-                        "score": [],
-                        "segment-text-start": [],
-                        "segment-text-end": []
-                    }
-                    wdx = 0
-                    cdx_prev = cdx + 2
-                    sub_seg_idx += 1
-                    seg_start_actual, seg_end_actual = None, None
-
-
-                # take min-max for actual segment-level timestamp
-                if seg_start_actual is None and start is not None:
-                    seg_start_actual = start
-                if end is not None:
-                    seg_end_actual = end
-
-
-            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
-        
-        start = interpolate_nans(pd.DataFrame(aligned_subsegments)["start"], method=interpolate_method) 
-        end = interpolate_nans(pd.DataFrame(aligned_subsegments)["end"], method=interpolate_method)
-        for idx, seg in enumerate(aligned_subsegments):
-            seg['start'] = start.iloc[idx]
-            seg['end'] = end.iloc[idx]
-
-        aligned_segments += aligned_subsegments
-
-    # create word level segments for .srt
-    word_seg = []
-    for seg in aligned_segments:
-        if model_lang in LANGUAGES_WITHOUT_SPACES:
-            # character based
-            seg["word-segments"] = seg["char-segments"]
-            seg["word-segments"]["segment-text-start"] = range(len(seg['word-segments']['start']))
-            seg["word-segments"]["segment-text-end"] = range(1, len(seg['word-segments']['start'])+1)
-
-        wseg = pd.DataFrame(seg["word-segments"]).replace({np.nan:None})
-        for wdx, wrow in wseg.iterrows():
-            if wrow["start"] is not None:
-                word_seg.append(
-                    {
-                        "start": wrow["start"],
-                        "end": wrow["end"],
-                        "text": seg["text"][int(wrow["segment-text-start"]):int(wrow["segment-text-end"])]
-                    }
-                )
-
-    return {"segments": aligned_segments, "word_segments": word_seg}
-
-def load_align_model(language_code, device, model_name=None):
-    if model_name is None:
-        # use default model
-        if language_code in DEFAULT_ALIGN_MODELS_TORCH:
-            model_name = DEFAULT_ALIGN_MODELS_TORCH[language_code]
-        elif language_code in DEFAULT_ALIGN_MODELS_HF:
-            model_name = DEFAULT_ALIGN_MODELS_HF[language_code]
-        else:
-            print(f"There is no default alignment model set for this language ({language_code}).\
-                Please find a wav2vec2.0 model finetuned on this language in https://huggingface.co/models, then pass the model name in --align_model [MODEL_NAME]")
-            raise ValueError(f"No default align-model for language: {language_code}")
-
-    if model_name in torchaudio.pipelines.__all__:
-        pipeline_type = "torchaudio"
-        bundle = torchaudio.pipelines.__dict__[model_name]
-        align_model = bundle.get_model().to(device)
-        labels = bundle.get_labels()
-        align_dictionary = {c.lower(): i for i, c in enumerate(labels)}
-    else:
-        try:
-            processor = AutoProcessor.from_pretrained(model_name)
-            align_model = Wav2Vec2ForCTC.from_pretrained(model_name)
-        except Exception as e:
-            print(e)
-            print(f"Error loading model from huggingface, check https://huggingface.co/models for finetuned wav2vec2.0 models")
-            raise ValueError(f'The chosen align_model "{model_name}" could not be found in huggingface (https://huggingface.co/models) or torchaudio (https://pytorch.org/audio/stable/pipelines.html#id14)')
-        pipeline_type = "huggingface"
-        align_model = align_model.to(device)
-        labels = processor.tokenizer.get_vocab()
-        align_dictionary = {char.lower(): code for char,code in processor.tokenizer.get_vocab().items()}
-
-    align_metadata = {"language": language_code, "dictionary": align_dictionary, "type": pipeline_type}
-
-    return align_model, align_metadata
-
-
 def merge_chunks(segments, chunk_size=CHUNK_LENGTH):
     """
-    Merge VAD segments into larger segments of 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_start = 0
     curr_end = 0
     merged_segments = []
     seg_idxs = []
     speaker_idxs = []
-    for sdx, seg in enumerate(segments):
+
+    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,
@@ -668,12 +318,9 @@ def transcribe_with_vad(
     prev = 0
     output = {"segments": []}
 
-    vad_segments_list = []
     vad_segments = vad_pipeline(audio)
-    for speech_turn in vad_segments.get_timeline().support():
-        vad_segments_list.append(Segment(speech_turn.start, speech_turn.end, "UNKNOWN"))
     # merge segments to approx 30s inputs to make whisper most appropraite
-    vad_segments = merge_chunks(vad_segments_list)
+    vad_segments = merge_chunks(vad_segments)
 
     for sdx, seg_t in enumerate(vad_segments):
         if verbose:
@@ -702,56 +349,223 @@ def transcribe_with_vad(
     return output
 
 
-def assign_word_speakers(diarize_df, result_segments, fill_nearest=False):
+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
+    """
 
-    for seg in result_segments:
+    if mel is None:
-        wdf = pd.DataFrame(seg['word-segments'])
+        mel = log_mel_spectrogram(audio)
-        if len(wdf['start'].dropna()) == 0:
-            wdf['start'] = seg['start']
-            wdf['end'] = seg['end']
-        speakers = []
-        for wdx, wrow in wdf.iterrows():
-            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]
-            speakers.append(speaker)
-        seg['word-segments']['speaker'] = speakers
-        seg["speaker"] = pd.Series(speakers).value_counts().index[0]
     
-    # create word level segments for .srt
+    vad_segments = vad_pipeline(audio)
-    word_seg = []
+    # merge segments to approx 30s inputs to make whisper most appropraite
-    for seg in result_segments:
+    vad_segments = merge_chunks(vad_segments)
-        wseg = pd.DataFrame(seg["word-segments"])
+
-        for wdx, wrow in wseg.iterrows():
+    ################################
-            if wrow["start"] is not None:
+    ### START of parallelization ###
-                speaker = wrow['speaker']
+    ################################
-                if speaker is None or speaker == np.nan:
+
-                    speaker = "UNKNOWN"
+    # pad mel to a same length
-                word_seg.append(
+    start_seconds = [i['start'] for i in vad_segments]
-                    {
+    end_seconds = [i['end'] for i in vad_segments]
-                        "start": wrow["start"],
+    duration_list = np.array(end_seconds) - np.array(start_seconds)
-                        "end": wrow["end"],
+    max_length = round(30 / (HOP_LENGTH / SAMPLE_RATE))
-                        "text": f"[{speaker}]: " + seg["text"][int(wrow["segment-text-start"]):int(wrow["segment-text-end"])]
+    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
 
-    # TODO: create segments but split words on new speaker
+            add_segment(
+                start=timestamp_offset,
+                end=timestamp_offset + duration,
+                text_tokens=tokens,
+                result=result,
+            )
 
-    return result_segments, word_seg
+            seek += segment_shape
+            all_tokens.extend(tokens.tolist())
 
-class Segment:
+        result = dict(text=tokenizer.decode(all_tokens), segments=all_segments, language=language)
-    def __init__(self, start, end, speaker=None):
+        output["segments"].append(
-        self.start = start
+            {
-        self.end = end
+                "start": seg_t["start"],
-        self.speaker = speaker
+                "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():
@@ -769,14 +583,14 @@ def cli():
     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("--vad_input", default=None, type=str)
+    parser.add_argument("--parallel_bs", default=-1, type=int, help="Enable parallel transcribing if > 1")
     # diarization params
-    parser.add_argument("--diarize", action='store_true')
+    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"], help="File type for desired output save")
+    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")
 
@@ -799,6 +613,7 @@ def cli():
     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")
@@ -812,24 +627,33 @@ def cli():
     align_from_prev: bool = args.pop("align_from_prev")
     interpolate_method: bool = args.pop("interpolate_method")
     
+    hf_token: str = args.pop("hf_token")
     vad_filter: bool = args.pop("vad_filter")
-    vad_input: bool = args.pop("vad_input")
+    parallel_bs: int = args.pop("parallel_bs")
 
     diarize: bool = args.pop("diarize")
     min_speakers: int = args.pop("min_speakers")
     max_speakers: int = args.pop("max_speakers")
 
     vad_pipeline = None
-    if vad_input is not None:
+    if vad_filter:
-        vad_input = pd.read_csv(vad_input, header=None, sep= " ")
+        if hf_token is None:
-    elif vad_filter:
+            print("Warning, no huggingface token used, needs to be saved in environment variable, otherwise will throw error loading VAD model...")
-        from pyannote.audio import Pipeline
+        from pyannote.audio import Inference
-        vad_pipeline = Pipeline.from_pretrained("pyannote/voice-activity-detection")
+        vad_pipeline = Inference(
+            "pyannote/segmentation",
+            pre_aggregation_hook=lambda segmentation: segmentation,
+            use_auth_token=hf_token,
+            device=torch.device(device),
+        )
 
     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")
+        diarize_pipeline = Pipeline.from_pretrained("pyannote/speaker-diarization@2.1",
+                                    use_auth_token=hf_token)
 
     os.makedirs(output_dir, exist_ok=True)
 
@@ -857,8 +681,12 @@ def cli():
 
     for audio_path in args.pop("audio"):
         if vad_filter:
-            print("Performing VAD...")
+            if parallel_bs > 1:
-            result = transcribe_with_vad(model, audio_path, vad_pipeline, temperature=temperature, **args)
+                print("Performing VAD and parallel transcribing ...")
+                result = transcribe_with_vad_parallel(model, audio_path, vad_pipeline, temperature=temperature, batch_size=parallel_bs, **args)
+            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)
@@ -868,6 +696,7 @@ def cli():
             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_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)
@@ -901,7 +730,7 @@ def cli():
 
         # save TSV
         if output_type in ["tsv", "all"]:
-            with open(os.path.join(output_dir, audio_basename + ".srt"), "w", encoding="utf-8") as srt:
+            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
@@ -915,10 +744,20 @@ def cli():
             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
+        # # save ASS character-level
-        if output_type in ["ass-char", "all"]:
+        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()

whisperx/utils.py

@@ -2,6 +2,7 @@ import os
 import zlib
 from typing import Callable, TextIO, Iterator, Tuple
 import pandas as pd
+import numpy as np
 
 def exact_div(x, y):
     assert x % y == 0
@@ -64,8 +65,8 @@ def write_vtt(transcript: Iterator[dict], file: TextIO):
 def write_tsv(transcript: Iterator[dict], file: TextIO):
     print("start", "end", "text", sep="\t", file=file)
     for segment in transcript:
-        print(round(1000 * segment['start']), file=file, end="\t")
+        print(segment['start'], file=file, end="\t")
-        print(round(1000 * segment['end']), file=file, end="\t")
+        print(segment['end'], file=file, end="\t")
         print(segment['text'].strip().replace("\t", " "), file=file, flush=True)
 
 
@@ -206,6 +207,8 @@ def write_ass(transcript: Iterator[dict],
     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']
@@ -214,7 +217,7 @@ def write_ass(transcript: Iterator[dict],
             else:
                 speaker_str = ""
             for cdx, crow in res_segs.iterrows():
-                if crow['start'] is not None:
+                if not np.isnan(crow['start']):
                     if resolution == "char":
                         idx_0 = cdx
                         idx_1 = cdx + 1

whisperx/vad.py

@@ -0,0 +1,185 @@
+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()