update whisperx, add support of setting vad_method - silero or pyannote

.gitignore

@@ -5,6 +5,9 @@ __pycache__/
 
 # Custom
 models
+compose-dev.yaml
+scripts
+whisperx
 
 # C extensions
 *.so

requirements.txt

@@ -2,7 +2,7 @@ fastapi==0.114.1
 uvicorn==0.30.6
 pydantic==2.9.1
 aiohttp==3.10.5
-whisperx @ git+https://github.com/Nyralei/whisperX.git@25c3fc61f795bf3fd40663098633fdba893eab0e
+whisperx @ git+https://github.com/Nyralei/whisperX.git@a44b97fc42d9f2a102cb70e5abcad65d9fceb85c
 ctranslate2==4.4.0
 numpy==1.26.4
 tqdm==4.66.5

src/whisperx_api_server/config.py

@@ -132,6 +132,10 @@ class Device(str, Enum):
     CUDA = "cuda"
     AUTO = "auto"
 
+class VadMethod(str, Enum):
+    SILERO = "silero"
+    PYANNOTE = "pyannote"
+
 class WhisperConfig(BaseModel):
     """See https://github.com/SYSTRAN/faster-whisper/blob/master/faster_whisper/transcribe.py#L599."""
 
@@ -148,6 +152,10 @@ class WhisperConfig(BaseModel):
     compute_type: Quantization = Field(default=Quantization.DEFAULT)
     cpu_threads: int = Field(default=0)
     num_workers: int = Field(default=1)
+    vad_method: VadMethod = Field(default=VadMethod.PYANNOTE)
+    vad_model: str = Field(default=None)
+    vad_options: dict = Field(default=None)
+
 
 class Config(BaseSettings):
     """

src/whisperx_api_server/transcriber.py

@@ -101,6 +101,9 @@ async def transcribe(
             compute_type=whispermodel.compute_type,
             language=language,
             asr_options=asr_options,
+            vad_model=config.whisper.vad_model,
+            vad_method=config.whisper.vad_method,
+            vad_options=config.whisper.vad_options,
             model=whispermodel,
         )
         logger.info(f"Loading model took {time.time() - model_loading_start:.2f} seconds")