Add max_single_segment_time

transcriber/AutoTranscriber.py

@@ -25,6 +25,7 @@ class AutoTranscriber:
         use_denoiser=False,
         with_punct=True,
         offset_in_seconds=-0.25,
+        max_length_seconds=5,
         sr=16000,
     ):
         self.corrector = corrector
@@ -32,9 +33,10 @@ class AutoTranscriber:
         self.with_punct = with_punct
         self.sr = sr
         self.offset_in_seconds = offset_in_seconds
+        self.max_length_seconds = max_length_seconds
 
         # Initialize models
-        self.vad_model = AutoModel(model="fsmn-vad", device=device,)
+        self.vad_model = AutoModel(model="fsmn-vad", device=device, max_single_segment_time=self.max_length_seconds * 1000)
         self.asr_model = AutoModel(
             model="iic/SenseVoiceSmall",
             vad_model=None,  # We'll handle VAD separately
@@ -64,14 +66,13 @@ class AutoTranscriber:
         #     speech, _ = denoiser(speech, sr)
 
         if sr != 16_000:
-            speech = resample(speech, sr, 16_000,
-                              filter="kaiser_best", parallel=True)
+            speech = resample(speech, sr, 16_000, filter="kaiser_best", parallel=True)
 
         # Get VAD segments
         logger.info("Segmenting speech...")
 
         start_time = time.time()
-        vad_results = self.vad_model.generate(input=speech)
+        vad_results = self.vad_model.generate(input=speech, disable_pbar=True)
         logger.info("VAD took %.2f seconds", time.time() - start_time)
 
         if not vad_results or not vad_results[0]["value"]: