websocket new version for offline 2pass send bytes

游雁

2023-05-13 9dad49c3a1f2495384bab4cc3763e4f8a461da00

 funasr/runtime/python/websocket/ws_server_online.py

@@ -26,74 +26,72 @@
print("model loading")

inference_pipeline_asr_online = pipeline(
    task=Tasks.auto_speech_recognition,
    model=args.asr_model_online,
    ngpu=args.ngpu,
    ncpu=args.ncpu,
    model_revision='v1.0.4')
   task=Tasks.auto_speech_recognition,
   model=args.asr_model_online,
   ngpu=args.ngpu,
   ncpu=args.ncpu,
   model_revision='v1.0.4')

print("model loaded")



async def ws_serve(websocket, path):
    frames_asr_online = []
    global websocket_users
    websocket_users.add(websocket)
    websocket.param_dict_asr_online = {"cache": dict()}
    print("new user connected",flush=True)
    try:
        async for message in websocket:
            
 
            if isinstance(message,str):
              messagejson = json.loads(message)
               
              if "is_speaking" in messagejson:
                  websocket.is_speaking = messagejson["is_speaking"]  
                  websocket.param_dict_asr_online["is_final"] = not websocket.is_speaking
              if "is_finished" in messagejson:
                  websocket.is_speaking = False
                  websocket.param_dict_asr_online["is_final"] = True
              if "chunk_interval" in messagejson:
                  websocket.chunk_interval=messagejson["chunk_interval"]
              if "wav_name" in messagejson:
                  websocket.wav_name = messagejson.get("wav_name", "demo")
              if "chunk_size" in messagejson:
                  websocket.param_dict_asr_online["chunk_size"] = messagejson["chunk_size"]
            # if has bytes in buffer or message is bytes
            if len(frames_asr_online)>0 or not isinstance(message,str):
               if not isinstance(message,str):
                 frames_asr_online.append(message)
               if len(frames_asr_online) % websocket.chunk_interval == 0 or not websocket.is_speaking:
                    audio_in = b"".join(frames_asr_online)
                    if not websocket.is_speaking:
                       #padding 0.5s at end gurantee that asr engine can fire out last word
                       audio_in=audio_in+b''.join(np.zeros(int(16000*0.5),dtype=np.int16))
                    await async_asr_online(websocket,audio_in)
                    frames_asr_online = []
   frames_asr_online = []
   global websocket_users
   websocket_users.add(websocket)
   websocket.param_dict_asr_online = {"cache": dict()}
   websocket.wav_name = "microphone"
   print("new user connected",flush=True)
   try:
      async for message in websocket:
			
			
         if isinstance(message, str):
            messagejson = json.loads(message)
				
            if "is_speaking" in messagejson:
               websocket.is_speaking = messagejson["is_speaking"]
               websocket.param_dict_asr_online["is_final"] = not websocket.is_speaking
            if "chunk_interval" in messagejson:
               websocket.chunk_interval=messagejson["chunk_interval"]
            if "wav_name" in messagejson:
               websocket.wav_name = messagejson.get("wav_name")
            if "chunk_size" in messagejson:
               websocket.param_dict_asr_online["chunk_size"] = messagejson["chunk_size"]
         # if has bytes in buffer or message is bytes
         if len(frames_asr_online) > 0 or not isinstance(message, str):
            if not isinstance(message,str):
               frames_asr_online.append(message)
            if len(frames_asr_online) % websocket.chunk_interval == 0 or not websocket.is_speaking:
               audio_in = b"".join(frames_asr_online)
               # if not websocket.is_speaking:
                  #padding 0.5s at end gurantee that asr engine can fire out last word
                  # audio_in=audio_in+b''.join(np.zeros(int(16000*0.5),dtype=np.int16))
               await async_asr_online(websocket,audio_in)
               frames_asr_online = []
	
	
   except websockets.ConnectionClosed:
      print("ConnectionClosed...", websocket_users)
      websocket_users.remove(websocket)
   except websockets.InvalidState:
      print("InvalidState...")
   except Exception as e:
      print("Exception:", e)

     
    except websockets.ConnectionClosed:
        print("ConnectionClosed...", websocket_users)
        websocket_users.remove(websocket)
    except websockets.InvalidState:
        print("InvalidState...")
    except Exception as e:
        print("Exception:", e)

 
async def async_asr_online(websocket,audio_in):
            if len(audio_in) > 0:
                audio_in = load_bytes(audio_in)
                rec_result = inference_pipeline_asr_online(audio_in=audio_in,
                                                           param_dict=websocket.param_dict_asr_online)
                if websocket.param_dict_asr_online["is_final"]:
                    websocket.param_dict_asr_online["cache"] = dict()
                if "text" in rec_result:
                    if rec_result["text"] != "sil" and rec_result["text"] != "waiting_for_more_voice":
                        message = json.dumps({"mode": "online", "text": rec_result["text"], "wav_name": websocket.wav_name})
                        await websocket.send(message)
   if len(audio_in) > 0:
      audio_in = load_bytes(audio_in)
      rec_result = inference_pipeline_asr_online(audio_in=audio_in,
                                                 param_dict=websocket.param_dict_asr_online)
      if websocket.param_dict_asr_online.get("is_final", False):
         websocket.param_dict_asr_online["cache"] = dict()
      if "text" in rec_result:
         if rec_result["text"] != "sil" and rec_result["text"] != "waiting_for_more_voice":
            message = json.dumps({"mode": "online", "text": rec_result["text"], "wav_name": websocket.wav_name})
            await websocket.send(message)