add translate

transcribe/whisper_llm_serve.py

@@ -116,6 +116,7 @@ class PywhisperInference:
     whisper_model = None
     llm_model = None
     vad_model = None
+    init_event = None
 
     @classmethod
     def initializer(cls, event:mp.Event, warmup=True):
@@ -133,7 +134,14 @@ class PywhisperInference:
         # init llamacpp
         cls.llm_model = QwenTranslator(config.LLM_MODEL_PATH, config.LLM_SYS_PROMPT)
         cls.vad_model = VoiceActivityDetector()
+        cls.init_event = event
         event.set()
+     
+    
+    @classmethod
+    def init(cls):
+        pass
+        
     
     @classmethod
     def warmup(cls, warmup_steps=1):
@@ -190,8 +198,8 @@ class PyWhiperCppServe(ServeClientBase):
         # 进程初始化后再开始收音频
         self._ready_state = mp.Event()
         self._pool = PPool(
-            max_workers=1, initializer=partial(PywhisperInference.initializer, event=self._ready_state))
-    
+            max_workers=1, initializer=PywhisperInference.initializer, initargs=(self._ready_state, ))
+        self._pool.submit(PywhisperInference.init_event)
         logger.info('Create a process to process audio.')
         self.send_ready_state()
         # self.load_frame_thread = threading.Thread(target=self.load_frame_from_queue)
@@ -203,8 +211,7 @@ class PyWhiperCppServe(ServeClientBase):
         self.trans_thread.start()
 
     def send_ready_state(self):
-        while not self._ready_state:
-            time.sleep(0.1)
+        self._ready_state.wait()
     
         self.websocket.send(json.dumps({
             "uid": self.client_uid,
@@ -238,7 +245,6 @@ class PyWhiperCppServe(ServeClientBase):
             else:
                 self.frames_np = np.append(self.frames_np,frame_np)
 
-    
 
     def update_audio_buffer(self, last_offset):
         with self.lock:
@@ -275,10 +281,9 @@ class PyWhiperCppServe(ServeClientBase):
             markers = config.PAUSE_END_MARKERS
             is_end = False
         else:
-            # 使用句号 问好等长句结尾符号作为判断
+            # 使用句号 长句结尾符号作为判断
             markers = config.SENTENCE_END_MARKERS
             is_end = True
-
     
         for idx, seg in enumerate(segments):
             left_watch_sequences.append(seg)
@@ -287,7 +292,7 @@ class PyWhiperCppServe(ServeClientBase):
                 rest_buffer_duration = (len(audio_buffer) - seg_index) / self.sample_rate
                 # is_end = any(i in seg.text for i  in config.SENTENCE_END_MARKERS)
                 right_watch_sequences = segments[min(idx+1, len(segments)):]
-                if rest_buffer_duration >= 1.5:
+                if rest_buffer_duration >= 1:
                     left_watch_idx = seg_index
                 break
         return left_watch_idx, left_watch_sequences, right_watch_sequences, is_end
@@ -321,8 +326,6 @@ class PyWhiperCppServe(ServeClientBase):
                 continue
 
             audio_buffer = self.get_audio_chunk_for_processing()
-            logger.info(f"Processing audio with duration: {len(audio_buffer)}")
-            # segments = self.transcribe_audio(audio_buffer)
             # c+= 1
             # name = f"dev-{c}.wav"
             # save_to_wave(name, audio_buffer)
@@ -362,7 +365,7 @@ class PyWhiperCppServe(ServeClientBase):
                 context=message,
                 from_=self.language,
                 to=self.dst_lang,
-                tran_content="this is english sample",
+                tran_content=self.translate_text(message),
                 partial=False
             )
             if self._segment_manager.string.strip():
@@ -372,7 +375,7 @@ class PyWhiperCppServe(ServeClientBase):
                     context=self._segment_manager.string,
                     from_=self.language,
                     to=self.dst_lang,
-                    tran_content="this is english sample",
+                    tran_content=self.translate_text(self._segment_manager.string.stri()),
             )
         
         else:
@@ -384,7 +387,7 @@ class PyWhiperCppServe(ServeClientBase):
                 context=message,
                 from_=self.language,
                 to=self.dst_lang,
-                tran_content="this is english sample",
+                tran_content=self.translate_text(message),
             )
     
     def send_to_client(self, data:TransResult):