funasr-onnx http

游雁

2023-11-10 d83e999f599fded04f72483fd24532e864511c05

funasr-onnx http

 runtime/python/onnxruntime/funasr_client_http.py

New file
@@ -0,0 +1,18 @@
import base64

import requests

import threading





with open('A2_0.wav','rb') as f:

    test_wav_bytes=f.read()

url='http://127.0.0.1:8888/api/asr'

def send_post(i,url,wav_bytes):

    r1=requests.post(url,json={'wav_base64':str(base64.b64encode(wav_bytes), 'utf-8')})

    print('线程:',i,r1.json())





for i in range(100):

    t = threading.Thread(target=send_post, args=(i,url,test_wav_bytes,))

    t.start()

    #t.join()

print('完成测试')

 runtime/python/onnxruntime/funasr_server_http.py

New file
@@ -0,0 +1,38 @@
import argparse

import base64

import io

import soundfile as sf

import uvicorn

from fastapi import FastAPI, Body

app = FastAPI()



from funasr_onnx import Paraformer

model_dir = "damo/speech_paraformer-large_asr_nat-zh-cn-16k-common-vocab8404-onnx"

model = Paraformer(model_dir, batch_size=1, quantize=True)



async def recognition_onnx(waveform):

    result=model(waveform)[0]["preds"][0]

    return result

@app.post("/api/asr")

async def asr(item: dict = Body(...)):

    try:

        audio_bytes= base64.b64decode(bytes(item['wav_base64'], 'utf-8'))

        waveform, _ = sf.read(io.BytesIO(audio_bytes))

        result=await recognition_onnx(waveform)

        ret = {"results": result, "code": 0}

    except:

        print('请求出错，这里是处理出错的')

        ret = {"results": '', "code": 1}

    return ret





if __name__ == '__main__':

    parser = argparse.ArgumentParser(description='API Service')

    parser.add_argument('--listen', default='0.0.0.0', type=str, help='the network to listen')

    parser.add_argument('--port', default=8888, type=int, help='the port to listen')

    args = parser.parse_args()



    print('start...')

    print('server on:',args)

    
    uvicorn.run(app, host=args.listen, port=args.port)