Add vad_threshold to Context

whispering/schema.py

@@ -41,6 +41,7 @@ class Context(BaseModel, arbitrary_types_allowed=True):
     logprob_threshold: Optional[float] = -1.0
     compression_ratio_threshold: Optional[float] = 2.4
     buffer_threshold: Optional[float] = 0.5
+    vad_threshold: float = 0.5
 
 
 class ParsedChunk(BaseModel):

whispering/transcriber.py

@@ -240,6 +240,7 @@ class WhisperStreamingTranscriber:
                 for v in self.vad(
                     audio=audio,
                     total_block_number=1,
+                    threshold=ctx.vad_threshold,
                 )
             ]
             if len(x) == 0:  # No speech

whispering/vad.py

@@ -22,7 +22,7 @@ class VAD:
         self,
         *,
         audio: np.ndarray,
-        threshold: float = 0.5,
+        threshold: float,
         total_block_number: Optional[int] = None,
     ) -> Iterator[SpeechSegment]:
         # audio.shape should be multiple of (N_FRAMES,)