websocket

游雁

2023-03-23 3dd708947e2939f836c6f85dff69ff089095a2db

 funasr/runtime/python/websocket/ASR_server.py

@@ -1,4 +1,10 @@
# server.py   注意本例仅处理单个clent发送的语音数据，并未对多client连接进行判断和处理
import asyncio
import websockets
import time
from queue import Queue
import threading
import argparse

from modelscope.pipelines import pipeline
from modelscope.utils.constant import Tasks
from modelscope.utils.logger import get_logger
@@ -7,12 +13,6 @@
logger = get_logger(log_level=logging.CRITICAL)
logger.setLevel(logging.CRITICAL)

import asyncio
import websockets
import time
from queue import Queue
import threading
import argparse

parser = argparse.ArgumentParser()
parser.add_argument("--host",
@@ -36,7 +36,7 @@

parser.add_argument("--punc_model",
                    type=str,
                    default="",
                    default="damo/punc_ct-transformer_zh-cn-common-vad_realtime-vocab272727",
                    help="model from modelscope")
parser.add_argument("--ngpu",
                    type=int,
@@ -49,25 +49,35 @@
voices = Queue()
speek = Queue()

# 创建一个VAD对象
vad_pipline = pipeline(
# vad
inference_pipeline_vad = pipeline(
    task=Tasks.voice_activity_detection,
    model=args.vad_model,
    model_revision="v1.2.0",
    output_dir=None,
    batch_size=1,
    mode='online'
    mode='online',
    ngpu=args.ngpu,
)
param_dict_vad = {'in_cache': dict(), "is_final": False}
  
# 创建一个ASR对象
param_dict = dict()
# asr
param_dict_asr = dict()
# param_dict["hotword"] = "小五 小五月"  # 设置热词，用空格隔开
inference_pipeline2 = pipeline(
inference_pipeline_asr = pipeline(
    task=Tasks.auto_speech_recognition,
    model=args.asr_model,
    param_dict=param_dict,
    param_dict=param_dict_asr,
    ngpu=args.ngpu,
)

inference_pipline_punc = pipeline(
    task=Tasks.punctuation,
    model=args.punc_model,
    model_revision="v1.0.1",
    ngpu=args.ngpu,
)

print("model loaded")


@@ -90,7 +100,7 @@
    global vad_pipline, param_dict_vad
    #print(type(data))
    # print(param_dict_vad)
    segments_result = vad_pipline(audio_in=data, param_dict=param_dict_vad)
    segments_result = inference_pipeline_vad(audio_in=data, param_dict=param_dict_vad)
    # print(segments_result)
    # print(param_dict_vad)
    speech_start = False
@@ -111,7 +121,7 @@
        while not speek.empty():
            audio_in = speek.get()
            speek.task_done()
            rec_result = inference_pipeline2(audio_in=audio_in)
            rec_result = inference_pipeline_asr(audio_in=audio_in)
            print(rec_result)
            time.sleep(0.1)
        time.sleep(0.1)