[fix]: update vad threshold.

main.py

@@ -57,16 +57,17 @@ async def root():
 async def translate(websocket: WebSocket):
     query_parameters_dict = websocket.query_params
     from_lang, to_lang = query_parameters_dict.get('from'), query_parameters_dict.get('to')
-    
+
     client = WhisperTranscriptionService(
         websocket,
         pipe,
-        language="en",
+        language=from_lang,
+        dst_lang=to_lang,
         client_uid=f"{uuid1()}",
     )
 
 
-    if from_lang and to_lang:
+    if from_lang and to_lang and client:
         client.set_language(from_lang, to_lang)
         logger.info(f"Source lange: {from_lang}  -> Dst lange: {to_lang}")
         await websocket.accept()
@@ -75,7 +76,7 @@ async def translate(websocket: WebSocket):
             frame_data = await get_audio_from_websocket(websocket)
             client.add_frames(frame_data)
     except WebSocketDisconnect:
-        return 
+        return
 
 if __name__ == '__main__':
     freeze_support()

transcribe/helpers/vadprocessor.py

@@ -278,12 +278,12 @@ class VadProcessor:
             cache_s=0.15,
             sr=16000
     ):
-        self.prob_thres = prob_threshold
+        self.prob_threshold = prob_threshold
         self.cache_s = cache_s
         self.sr = sr
         self.silence_s = silence_s
 
-        self.vad = VadV2(self.prob_thres, self.sr, self.silence_s * 1000, self.cache_s * 1000, max_speech_duration_s=15)
+        self.vad = VadV2(self.prob_threshold, self.sr, self.silence_s * 1000, self.cache_s * 1000, max_speech_duration_s=15)
 
 
     def process_audio(self, audio_buffer: np.ndarray):

transcribe/whisper_llm_serve.py

@@ -30,7 +30,8 @@ class WhisperTranscriptionService:
     DISCONNECT = "DISCONNECT"
 
     def __init__(self, websocket, pipe: TranslatePipes, language=None, dst_lang=None, client_uid=None):
-       
+        print('>>>>>>>>>>>>>>>> init service >>>>>>>>>>>>>>>>>>>>>>')
+        print('src_lang:', language)
         self.source_language = language  # 源语言
         self.target_language = dst_lang  # 目标翻译语言
         self.client_uid = client_uid
@@ -40,7 +41,7 @@ class WhisperTranscriptionService:
 
         # 音频处理相关
         self.sample_rate = 16000
- 
+
         self.lock = threading.Lock()
         self._frame_queue = queue.Queue()
         self._vad_frame_queue = queue.Queue()
@@ -49,7 +50,7 @@ class WhisperTranscriptionService:
         self.text_separator = self._get_text_separator(language)
         self.loop = asyncio.get_event_loop()
         # 发送就绪状态
-        
+
         self._transcrible_analysis = None
         # 启动处理线程
         self._translate_thread_stop = threading.Event()
@@ -57,7 +58,10 @@ class WhisperTranscriptionService:
 
         self.translate_thread = self._start_thread(self._transcription_processing_loop)
         self.frame_processing_thread = self._start_thread(self._frame_processing_loop)
-        self._vad = VadProcessor()
+        if language == "zh":
+            self._vad = VadProcessor(prob_threshold=0.8, silence_s=0.2, cache_s=0.15)
+        else:
+            self._vad = VadProcessor(prob_threshold=0.7, silence_s=0.2, cache_s=0.15)
         self.row_number = 0
         # for test
         self._transcrible_time_cost = 0.
@@ -66,9 +70,9 @@ class WhisperTranscriptionService:
             self._test_task_stop = threading.Event()
             self._test_queue = queue.Queue()
             self._test_thread = self._start_thread(self.test_data_loop)
-          
+
         # self._c = 0
-    
+
     def test_data_loop(self):
         writer = TestDataWriter()
         while not self._test_task_stop.is_set():
@@ -179,7 +183,7 @@ class WhisperTranscriptionService:
             if audio_buffer is None or len(audio_buffer) < int(self.sample_rate):
                 time.sleep(0.2)
                 continue
-                
+
             logger.debug(f"audio buffer size: {len(audio_buffer) / self.sample_rate:.2f}s")
             # try:
             meta_item = self._transcribe_audio(audio_buffer)