python-websocket funasr1.0 (#1310)

zhifu gao

2024-01-26 4f224c88068b66bcb6f81570da59d99c9bba8288

 runtime/python/websocket/funasr_wss_server.py

@@ -7,14 +7,7 @@
import numpy as np
import argparse
import ssl
from modelscope.pipelines import pipeline
from modelscope.utils.constant import Tasks
from modelscope.utils.logger import get_logger

tracemalloc.start()

logger = get_logger(log_level=logging.CRITICAL)
logger.setLevel(logging.CRITICAL)

parser = argparse.ArgumentParser()
parser.add_argument("--host",
@@ -29,24 +22,44 @@
                    help="grpc server port")
parser.add_argument("--asr_model",
                    type=str,
                    default="damo/speech_paraformer-large_asr_nat-zh-cn-16k-common-vocab8404-pytorch",
                    default="iic/speech_paraformer-large_asr_nat-zh-cn-16k-common-vocab8404-pytorch",
                    help="model from modelscope")
parser.add_argument("--asr_model_revision",
                    type=str,
                    default="v2.0.4",
                    help="")
parser.add_argument("--asr_model_online",
                    type=str,
                    default="damo/speech_paraformer-large_asr_nat-zh-cn-16k-common-vocab8404-online",
                    default="iic/speech_paraformer-large_asr_nat-zh-cn-16k-common-vocab8404-online",
                    help="model from modelscope")
parser.add_argument("--asr_model_online_revision",
                    type=str,
                    default="v2.0.4",
                    help="")
parser.add_argument("--vad_model",
                    type=str,
                    default="damo/speech_fsmn_vad_zh-cn-16k-common-pytorch",
                    default="iic/speech_fsmn_vad_zh-cn-16k-common-pytorch",
                    help="model from modelscope")
parser.add_argument("--vad_model_revision",
                    type=str,
                    default="v2.0.4",
                    help="")
parser.add_argument("--punc_model",
                    type=str,
                    default="damo/punc_ct-transformer_zh-cn-common-vad_realtime-vocab272727",
                    default="iic/punc_ct-transformer_zh-cn-common-vad_realtime-vocab272727",
                    help="model from modelscope")
parser.add_argument("--punc_model_revision",
                    type=str,
                    default="v2.0.4",
                    help="")
parser.add_argument("--ngpu",
                    type=int,
                    default=1,
                    help="0 for cpu, 1 for gpu")
parser.add_argument("--device",
                    type=str,
                    default="cuda",
                    help="cuda, cpu")
parser.add_argument("--ncpu",
                    type=int,
                    default=4,
@@ -68,213 +81,232 @@
websocket_users = set()

print("model loading")
from funasr import AutoModel

# asr
inference_pipeline_asr = pipeline(
    task=Tasks.auto_speech_recognition,
    model=args.asr_model,
    ngpu=args.ngpu,
    ncpu=args.ncpu,
    model_revision=None)


model_asr = AutoModel(model=args.asr_model,
                      model_revision=args.asr_model_revision,
                      ngpu=args.ngpu,
                      ncpu=args.ncpu,
                      device=args.device,
                      disable_pbar=True,
                      disable_log=True,
                      )
# asr
model_asr_streaming = AutoModel(model=args.asr_model_online,
                                model_revision=args.asr_model_online_revision,
                                ngpu=args.ngpu,
                                ncpu=args.ncpu,
                                device=args.device,
                                disable_pbar=True,
                                disable_log=True,
                                )
# vad
inference_pipeline_vad = pipeline(
    task=Tasks.voice_activity_detection,
    model=args.vad_model,
    model_revision=None,
    mode='online',
    ngpu=args.ngpu,
    ncpu=args.ncpu,
)
model_vad = AutoModel(model=args.vad_model,
                      model_revision=args.vad_model_revision,
                      ngpu=args.ngpu,
                      ncpu=args.ncpu,
                      device=args.device,
                      disable_pbar=True,
                      disable_log=True,
                      # chunk_size=60,
                      )

if args.punc_model != "":
    inference_pipeline_punc = pipeline(
        task=Tasks.punctuation,
        model=args.punc_model,
        model_revision="v1.0.2",
        ngpu=args.ngpu,
        ncpu=args.ncpu,
    )
   model_punc = AutoModel(model=args.punc_model,
                          model_revision=args.punc_model_revision,
                          ngpu=args.ngpu,
                          ncpu=args.ncpu,
                          device=args.device,
                          disable_pbar=True,
                           disable_log=True,
                          )
else:
    inference_pipeline_punc = None
   model_punc = None

inference_pipeline_asr_online = pipeline(
    task=Tasks.auto_speech_recognition,
    model=args.asr_model_online,
    ngpu=args.ngpu,
    ncpu=args.ncpu,
    model_revision='v1.0.7',
    update_model='v1.0.7',
    mode='paraformer_streaming')


print("model loaded! only support one client at the same time now!!!!")

async def ws_reset(websocket):
    print("ws reset now, total num is ",len(websocket_users))
    websocket.param_dict_asr_online = {"cache": dict()}
    websocket.param_dict_vad = {'in_cache': dict(), "is_final": True}
    websocket.param_dict_asr_online["is_final"]=True
    # audio_in=b''.join(np.zeros(int(16000),dtype=np.int16))
    # inference_pipeline_vad(audio_in=audio_in, param_dict=websocket.param_dict_vad)
    # inference_pipeline_asr_online(audio_in=audio_in, param_dict=websocket.param_dict_asr_online)
    await websocket.close()
    
    
   print("ws reset now, total num is ",len(websocket_users))

   websocket.status_dict_asr_online["cache"] = {}
   websocket.status_dict_asr_online["is_final"] = True
   websocket.status_dict_vad["cache"] = {}
   websocket.status_dict_vad["is_final"] = True
   websocket.status_dict_punc["cache"] = {}
	
   await websocket.close()


async def clear_websocket():
   for websocket in websocket_users:
       await ws_reset(websocket)
   websocket_users.clear()
 
 
       
   for websocket in websocket_users:
      await ws_reset(websocket)
   websocket_users.clear()



async def ws_serve(websocket, path):
    frames = []
    frames_asr = []
    frames_asr_online = []
    global websocket_users
    await clear_websocket()
    websocket_users.add(websocket)
    websocket.param_dict_asr = {}
    websocket.param_dict_asr_online = {"cache": dict()}
    websocket.param_dict_vad = {'in_cache': dict(), "is_final": False}
    websocket.param_dict_punc = {'cache': list()}
    websocket.vad_pre_idx = 0
    speech_start = False
    speech_end_i = -1
    websocket.wav_name = "microphone"
    websocket.mode = "2pass"
    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 "encoder_chunk_look_back" in messagejson:
                    websocket.param_dict_asr_online["encoder_chunk_look_back"] = messagejson["encoder_chunk_look_back"]
                if "decoder_chunk_look_back" in messagejson:
                    websocket.param_dict_asr_online["decoder_chunk_look_back"] = messagejson["decoder_chunk_look_back"]
                if "mode" in messagejson:
                    websocket.mode = messagejson["mode"]
            if len(frames_asr_online) > 0 or len(frames_asr) > 0 or not isinstance(message, str):
                if not isinstance(message, str):
                    frames.append(message)
                    duration_ms = len(message)//32
                    websocket.vad_pre_idx += duration_ms
        
                    # asr online
                    frames_asr_online.append(message)
                    websocket.param_dict_asr_online["is_final"] = speech_end_i != -1
                    if len(frames_asr_online) % websocket.chunk_interval == 0 or websocket.param_dict_asr_online["is_final"]:
                        if websocket.mode == "2pass" or websocket.mode == "online":
                            audio_in = b"".join(frames_asr_online)
                            await async_asr_online(websocket, audio_in)
                        frames_asr_online = []
                    if speech_start:
                        frames_asr.append(message)
                    # vad online
                    speech_start_i, speech_end_i = await async_vad(websocket, message)
                    if speech_start_i != -1:
                        speech_start = True
                        beg_bias = (websocket.vad_pre_idx-speech_start_i)//duration_ms
                        frames_pre = frames[-beg_bias:]
                        frames_asr = []
                        frames_asr.extend(frames_pre)
                # asr punc offline
                if speech_end_i != -1 or not websocket.is_speaking:
                    # print("vad end point")
                    if websocket.mode == "2pass" or websocket.mode == "offline":
                        audio_in = b"".join(frames_asr)
                        await async_asr(websocket, audio_in)
                    frames_asr = []
                    speech_start = False
                    # frames_asr_online = []
                    # websocket.param_dict_asr_online = {"cache": dict()}
                    if not websocket.is_speaking:
                        websocket.vad_pre_idx = 0
                        frames = []
                        websocket.param_dict_vad = {'in_cache': dict()}
                    else:
                        frames = frames[-20:]

     
    except websockets.ConnectionClosed:
        print("ConnectionClosed...", websocket_users,flush=True)
        await ws_reset(websocket)
        websocket_users.remove(websocket)
    except websockets.InvalidState:
        print("InvalidState...")
    except Exception as e:
        print("Exception:", e)
   frames = []
   frames_asr = []
   frames_asr_online = []
   global websocket_users
   await clear_websocket()
   websocket_users.add(websocket)
   websocket.status_dict_asr = {}
   websocket.status_dict_asr_online = {"cache": {}, "is_final": False}
   websocket.status_dict_vad = {'cache': {}, "is_final": False}
   websocket.status_dict_punc = {'cache': {}}
   websocket.chunk_interval = 10
   websocket.vad_pre_idx = 0
   speech_start = False
   speech_end_i = -1
   websocket.wav_name = "microphone"
   websocket.mode = "2pass"
   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.status_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.status_dict_asr_online["chunk_size"] = messagejson["chunk_size"]
            if "encoder_chunk_look_back" in messagejson:
               websocket.status_dict_asr_online["encoder_chunk_look_back"] = messagejson["encoder_chunk_look_back"]
            if "decoder_chunk_look_back" in messagejson:
               websocket.status_dict_asr_online["decoder_chunk_look_back"] = messagejson["decoder_chunk_look_back"]
            if "hotword" in messagejson:
               websocket.status_dict_asr["hotword"] = messagejson["hotword"]
            if "mode" in messagejson:
               websocket.mode = messagejson["mode"]
			
         websocket.status_dict_vad["chunk_size"] = int(websocket.status_dict_asr_online["chunk_size"][1]*60/websocket.chunk_interval)
         if len(frames_asr_online) > 0 or len(frames_asr) > 0 or not isinstance(message, str):
            if not isinstance(message, str):
               frames.append(message)
               duration_ms = len(message)//32
               websocket.vad_pre_idx += duration_ms
					
               # asr online
               frames_asr_online.append(message)
               websocket.status_dict_asr_online["is_final"] = speech_end_i != -1
               if len(frames_asr_online) % websocket.chunk_interval == 0 or websocket.status_dict_asr_online["is_final"]:
                  if websocket.mode == "2pass" or websocket.mode == "online":
                     audio_in = b"".join(frames_asr_online)
                     try:
                        await async_asr_online(websocket, audio_in)
                     except:
                        print(f"error in asr streaming, {websocket.status_dict_asr_online}")
                  frames_asr_online = []
               if speech_start:
                  frames_asr.append(message)
               # vad online
               try:
                  speech_start_i, speech_end_i = await async_vad(websocket, message)
               except:
                  print("error in vad")
               if speech_start_i != -1:
                  speech_start = True
                  beg_bias = (websocket.vad_pre_idx-speech_start_i)//duration_ms
                  frames_pre = frames[-beg_bias:]
                  frames_asr = []
                  frames_asr.extend(frames_pre)
            # asr punc offline
            if speech_end_i != -1 or not websocket.is_speaking:
               # print("vad end point")
               if websocket.mode == "2pass" or websocket.mode == "offline":
                  audio_in = b"".join(frames_asr)
                  try:
                     await async_asr(websocket, audio_in)
                  except:
                     print("error in asr offline")
               frames_asr = []
               speech_start = False
               frames_asr_online = []
               websocket.status_dict_asr_online["cache"] = {}
               if not websocket.is_speaking:
                  websocket.vad_pre_idx = 0
                  frames = []
                  websocket.status_dict_vad["cache"] = {}
               else:
                  frames = frames[-20:]
	
	
   except websockets.ConnectionClosed:
      print("ConnectionClosed...", websocket_users,flush=True)
      await ws_reset(websocket)
      websocket_users.remove(websocket)
   except websockets.InvalidState:
      print("InvalidState...")
   except Exception as e:
      print("Exception:", e)


async def async_vad(websocket, audio_in):

    segments_result = inference_pipeline_vad(audio_in=audio_in, param_dict=websocket.param_dict_vad)

    speech_start = -1
    speech_end = -1
    
    if len(segments_result) == 0 or len(segments_result["text"]) > 1:
        return speech_start, speech_end
    if segments_result["text"][0][0] != -1:
        speech_start = segments_result["text"][0][0]
    if segments_result["text"][0][1] != -1:
        speech_end = segments_result["text"][0][1]
    return speech_start, speech_end
	
   segments_result = model_vad.generate(input=audio_in, **websocket.status_dict_vad)[0]["value"]
   # print(segments_result)
	
   speech_start = -1
   speech_end = -1
	
   if len(segments_result) == 0 or len(segments_result) > 1:
      return speech_start, speech_end
   if segments_result[0][0] != -1:
      speech_start = segments_result[0][0]
   if segments_result[0][1] != -1:
      speech_end = segments_result[0][1]
   return speech_start, speech_end


async def async_asr(websocket, audio_in):
            if len(audio_in) > 0:
                # print(len(audio_in))
                rec_result = inference_pipeline_asr(audio_in=audio_in,
                                                    param_dict=websocket.param_dict_asr)
                # print(rec_result)
                if inference_pipeline_punc is not None and 'text' in rec_result and len(rec_result["text"])>0:
                    rec_result = inference_pipeline_punc(text_in=rec_result['text'],
                                                         param_dict=websocket.param_dict_punc)
                    # print("offline", rec_result)
                if 'text' in rec_result:
                    mode = "2pass-offline" if "2pass" in websocket.mode else websocket.mode
                    message = json.dumps({"mode": mode, "text": rec_result["text"], "wav_name": websocket.wav_name,"is_final":websocket.is_speaking})
                    await websocket.send(message)
   if len(audio_in) > 0:
      # print(len(audio_in))
      rec_result = model_asr.generate(input=audio_in, **websocket.status_dict_asr)[0]
      # print("offline_asr, ", rec_result)
      if model_punc is not None and len(rec_result["text"])>0:
         # print("offline, before punc", rec_result, "cache", websocket.status_dict_punc)
         rec_result = model_punc.generate(input=rec_result['text'], **websocket.status_dict_punc)[0]
         # print("offline, after punc", rec_result)
      if len(rec_result["text"])>0:
         # print("offline", rec_result)
         mode = "2pass-offline" if "2pass" in websocket.mode else websocket.mode
         message = json.dumps({"mode": mode, "text": rec_result["text"], "wav_name": websocket.wav_name,"is_final":websocket.is_speaking})
         await websocket.send(message)


async def async_asr_online(websocket, audio_in):
    if len(audio_in) > 0:
        # print(websocket.param_dict_asr_online.get("is_final", False))
        rec_result = inference_pipeline_asr_online(audio_in=audio_in,
                                                   param_dict=websocket.param_dict_asr_online)
        # print(rec_result)
        if websocket.mode == "2pass" and websocket.param_dict_asr_online.get("is_final", False):
            return
            #     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":
                # print("online", rec_result)
                mode = "2pass-online" if "2pass" in websocket.mode else websocket.mode
                message = json.dumps({"mode": mode, "text": rec_result["text"], "wav_name": websocket.wav_name,"is_final":websocket.is_speaking})
                await websocket.send(message)
   if len(audio_in) > 0:
      # print(websocket.status_dict_asr_online.get("is_final", False))
      rec_result = model_asr_streaming.generate(input=audio_in, **websocket.status_dict_asr_online)[0]
      # print("online, ", rec_result)
      if websocket.mode == "2pass" and websocket.status_dict_asr_online.get("is_final", False):
         return
         #     websocket.status_dict_asr_online["cache"] = dict()
      if len(rec_result["text"]):
         mode = "2pass-online" if "2pass" in websocket.mode else websocket.mode
         message = json.dumps({"mode": mode, "text": rec_result["text"], "wav_name": websocket.wav_name,"is_final":websocket.is_speaking})
         await websocket.send(message)

if len(args.certfile)>0:
    ssl_context = ssl.SSLContext(ssl.PROTOCOL_TLS_SERVER)
    
    # Generate with Lets Encrypt, copied to this location, chown to current user and 400 permissions
    ssl_cert = args.certfile
    ssl_key = args.keyfile
    
    ssl_context.load_cert_chain(ssl_cert, keyfile=ssl_key)
    start_server = websockets.serve(ws_serve, args.host, args.port, subprotocols=["binary"], ping_interval=None,ssl=ssl_context)
   ssl_context = ssl.SSLContext(ssl.PROTOCOL_TLS_SERVER)
	
   # Generate with Lets Encrypt, copied to this location, chown to current user and 400 permissions
   ssl_cert = args.certfile
   ssl_key = args.keyfile
	
   ssl_context.load_cert_chain(ssl_cert, keyfile=ssl_key)
   start_server = websockets.serve(ws_serve, args.host, args.port, subprotocols=["binary"], ping_interval=None,ssl=ssl_context)
else:
    start_server = websockets.serve(ws_serve, args.host, args.port, subprotocols=["binary"], ping_interval=None)
   start_server = websockets.serve(ws_serve, args.host, args.port, subprotocols=["binary"], ping_interval=None)
asyncio.get_event_loop().run_until_complete(start_server)
asyncio.get_event_loop().run_forever()