Merge c72c627d10 into d700b56c9c

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

@@ -17,6 +17,9 @@ jobs:
           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
 

.github/workflows/python-compatibility.yml

@@ -23,6 +23,9 @@ jobs:
           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
 

README.md

@@ -97,6 +97,25 @@ uv sync --all-extras --dev
 
 You may also need to install ffmpeg, rust etc. Follow openAI instructions here https://github.com/openai/whisper#setup.
 
+### Common Issues & Troubleshooting 🔧
+
+#### libcudnn Dependencies (GPU Users)
+
+If you're using WhisperX with GPU support and encounter errors like:
+
+- `Could not load library libcudnn_ops_infer.so.8`
+- `Unable to load any of {libcudnn_cnn.so.9.1.0, libcudnn_cnn.so.9.1, libcudnn_cnn.so.9, libcudnn_cnn.so}`
+- `libcudnn_ops_infer.so.8: cannot open shared object file: No such file or directory`
+
+This means your system is missing the CUDA Deep Neural Network library (cuDNN). This library is needed for GPU acceleration but isn't always installed by default.
+
+**Install cuDNN (example for apt based systems):**
+
+```bash
+sudo apt update
+sudo apt install libcudnn8 libcudnn8-dev -y
+```
+
 ### 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.)
@@ -170,7 +189,7 @@ result = model.transcribe(audio, batch_size=batch_size)
 print(result["segments"]) # before alignment
 
 # delete model if low on GPU resources
-# import gc; gc.collect(); torch.cuda.empty_cache(); del model
+# 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)
@@ -179,7 +198,7 @@ result = whisperx.align(result["segments"], model_a, metadata, audio, device, re
 print(result["segments"]) # after alignment
 
 # delete model if low on GPU resources
-# import gc; gc.collect(); torch.cuda.empty_cache(); del model_a
+# 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)

uv.lock

@@ -2787,7 +2787,7 @@ wheels = [
 
 [[package]]
 name = "whisperx"
-version = "3.3.3"
+version = "3.3.4"
 source = { editable = "." }
 dependencies = [
     { name = "ctranslate2" },

whisperx/__main__.py

@@ -43,6 +43,7 @@ def cli():
     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")

whisperx/alignment.py

@@ -5,7 +5,7 @@ C. Max Bain
 import math
 
 from dataclasses import dataclass
-from typing import Iterable, Optional, Union, List
+from typing import Iterable, Union, List, Callable, Optional
 
 import numpy as np
 import pandas as pd
@@ -120,6 +120,7 @@ def align(
     return_char_alignments: bool = False,
     print_progress: bool = False,
     combined_progress: bool = False,
+    on_progress: Callable[[int, int], None] = None
 ) -> AlignedTranscriptionResult:
     """
     Align phoneme recognition predictions to known transcription.
@@ -148,6 +149,9 @@ def align(
             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}%...")
+
+        if on_progress:
+            on_progress(sdx + 1, total_segments)
             
         num_leading = len(segment["text"]) - len(segment["text"].lstrip())
         num_trailing = len(segment["text"]) - len(segment["text"].rstrip())

whisperx/asr.py

@@ -1,6 +1,8 @@
 import os
-from typing import List, Optional, Union
 from dataclasses import replace
+import warnings
+from typing import List, Union, Optional, NamedTuple, Callable
+from enum import Enum
 
 import ctranslate2
 import faster_whisper
@@ -103,6 +105,12 @@ class FasterWhisperPipeline(Pipeline):
     # - add support for timestamp mode
     # - add support for custom inference kwargs
 
+    class TranscriptionState(Enum):
+        LOADING_AUDIO = "loading_audio"
+        GENERATING_VAD_SEGMENTS = "generating_vad_segments"
+        TRANSCRIBING = "transcribing"
+        FINISHED = "finished"
+
     def __init__(
         self,
         model: WhisperModel,
@@ -197,8 +205,12 @@ class FasterWhisperPipeline(Pipeline):
         print_progress=False,
         combined_progress=False,
         verbose=False,
+            on_progress: Callable[[TranscriptionState, Optional[int], Optional[int]], None] = None,
     ) -> TranscriptionResult:
         if isinstance(audio, str):
+            if on_progress:
+                on_progress(self.__class__.TranscriptionState.LOADING_AUDIO)
+
             audio = load_audio(audio)
 
         def data(audio, segments):
@@ -216,6 +228,8 @@ class FasterWhisperPipeline(Pipeline):
         else:
             waveform = Pyannote.preprocess_audio(audio)
             merge_chunks = Pyannote.merge_chunks
+        if on_progress:
+            on_progress(self.__class__.TranscriptionState.GENERATING_VAD_SEGMENTS)
 
         vad_segments = self.vad_model({"waveform": waveform, "sample_rate": SAMPLE_RATE})
         vad_segments = merge_chunks(
@@ -255,16 +269,22 @@ class FasterWhisperPipeline(Pipeline):
         segments: List[SingleSegment] = []
         batch_size = batch_size or self._batch_size
         total_segments = len(vad_segments)
+
+        if on_progress:
+            on_progress(self.__class__.TranscriptionState.TRANSCRIBING, 0, total_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}%...")
+
+            if on_progress:
+                on_progress(self.__class__.TranscriptionState.TRANSCRIBING, idx + 1, total_segments)
+
             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,
@@ -273,6 +293,9 @@ class FasterWhisperPipeline(Pipeline):
                 }
             )
 
+        if on_progress:
+            on_progress(self.__class__.TranscriptionState.FINISHED)
+
         # revert the tokenizer if multilingual inference is enabled
         if self.preset_language is None:
             self.tokenizer = None

whisperx/diarize.py

@@ -11,13 +11,14 @@ from whisperx.types import TranscriptionResult, AlignedTranscriptionResult
 class DiarizationPipeline:
     def __init__(
         self,
-        model_name="pyannote/speaker-diarization-3.1",
+        model_name=None,
         use_auth_token=None,
         device: Optional[Union[str, torch.device]] = "cpu",
     ):
         if isinstance(device, str):
             device = torch.device(device)
-        self.model = Pipeline.from_pretrained(model_name, use_auth_token=use_auth_token).to(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,

whisperx/transcribe.py

@@ -57,6 +57,7 @@ def transcribe_task(args: dict, parser: argparse.ArgumentParser):
     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")
 
     if args["language"] is not None:
@@ -204,8 +205,9 @@ def transcribe_task(args: dict, parser: argparse.ArgumentParser):
             )
         tmp_results = results
         print(">>Performing diarization...")
+        print(">>Using model:", diarize_model_name)
         results = []
-        diarize_model = DiarizationPipeline(use_auth_token=hf_token, device=device)
+        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