Dev gzf exp (#1654)

zhifu gao

2024-04-24 861147c7308b91068ffa02724fdf74ee623a909e

 runtime/python/websocket/funasr_wss_server.py

@@ -10,71 +10,56 @@


parser = argparse.ArgumentParser()
parser.add_argument("--host",
                    type=str,
                    default="0.0.0.0",
                    required=False,
                    help="host ip, localhost, 0.0.0.0")
parser.add_argument("--port",
                    type=int,
                    default=10095,
                    required=False,
                    help="grpc server port")
parser.add_argument("--asr_model",
parser.add_argument(
    "--host", type=str, default="0.0.0.0", required=False, help="host ip, localhost, 0.0.0.0"
)
parser.add_argument("--port", type=int, default=10095, required=False, help="grpc server port")
parser.add_argument(
    "--asr_model",
                    type=str,
                    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",
    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="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",
    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="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",
    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="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,
                    help="cpu cores")
parser.add_argument("--certfile",
    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, help="cpu cores")
parser.add_argument(
    "--certfile",
                    type=str,
                    default="../../ssl_key/server.crt",
                    required=False,
                    help="certfile for ssl")
    help="certfile for ssl",
)

parser.add_argument("--keyfile",
parser.add_argument(
    "--keyfile",
                    type=str,
                    default="../../ssl_key/server.key",
                    required=False,
                    help="keyfile for ssl")
    help="keyfile for ssl",
)
args = parser.parse_args()


@@ -84,7 +69,8 @@
from funasr import AutoModel

# asr
model_asr = AutoModel(model=args.asr_model,
model_asr = AutoModel(
    model=args.asr_model,
                      model_revision=args.asr_model_revision,
                      ngpu=args.ngpu,
                      ncpu=args.ncpu,
@@ -93,7 +79,8 @@
                      disable_log=True,
                      )
# asr
model_asr_streaming = AutoModel(model=args.asr_model_online,
model_asr_streaming = AutoModel(
    model=args.asr_model_online,
                                model_revision=args.asr_model_online_revision,
                                ngpu=args.ngpu,
                                ncpu=args.ncpu,
@@ -102,7 +89,8 @@
                                disable_log=True,
                                )
# vad
model_vad = AutoModel(model=args.vad_model,
model_vad = AutoModel(
    model=args.vad_model,
                      model_revision=args.vad_model_revision,
                      ngpu=args.ngpu,
                      ncpu=args.ncpu,
@@ -113,7 +101,8 @@
                      )

if args.punc_model != "":
   model_punc = AutoModel(model=args.punc_model,
    model_punc = AutoModel(
        model=args.punc_model,
                          model_revision=args.punc_model_revision,
                          ngpu=args.ngpu,
                          ncpu=args.ncpu,
@@ -125,8 +114,8 @@
   model_punc = None



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))
@@ -146,7 +135,6 @@
   websocket_users.clear()



async def ws_serve(websocket, path):
   frames = []
   frames_asr = []
@@ -156,8 +144,8 @@
   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.status_dict_vad = {"cache": {}, "is_final": False}
    websocket.status_dict_punc = {"cache": {}}
   websocket.chunk_interval = 10
   websocket.vad_pre_idx = 0
   speech_start = False
@@ -181,18 +169,24 @@
            if "chunk_size" in messagejson:
               chunk_size = messagejson["chunk_size"]
               if isinstance(chunk_size, str):
                  chunk_size = chunk_size.split(',')
                        chunk_size = chunk_size.split(",")
               websocket.status_dict_asr_online["chunk_size"] = [int(x) for x in chunk_size]
            if "encoder_chunk_look_back" in messagejson:
               websocket.status_dict_asr_online["encoder_chunk_look_back"] = messagejson["encoder_chunk_look_back"]
                    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"]
                    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)
            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)
@@ -202,7 +196,10 @@
               # 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 (
                        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:
@@ -243,7 +240,6 @@
               else:
                  frames = frames[-20:]
   
	
   except websockets.ConnectionClosed:
      print("ConnectionClosed...", websocket_users,flush=True)
      await ws_reset(websocket)
@@ -278,27 +274,46 @@
      # 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]
            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})
            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.status_dict_asr_online.get("is_final", False))
      rec_result = model_asr_streaming.generate(input=audio_in, **websocket.status_dict_asr_online)[0]
        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})
            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)
@@ -308,8 +323,12 @@
   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)
    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()