Python 使用VOSK进行语音识别

参考地址

• 安装vosk

  pip install vosk
  

• 下载示例代码

  git clone https://github.com/alphacep/vosk-api.git
  git clone https://gitee.com/dimld/vosk-api.git       # gitee资源
  

• 下载模型
  去官网下载预编译的模型文件，或者使用指令下载，主要是这个vosk-model-cn-0.15.zip，另外两个没用上

  wget -c https://alphacephei.com/vosk/models/vosk-model-small-cn-0.22.zip
  wget -c https://alphacephei.com/vosk/models/vosk-model-cn-0.15.zip
  wget -c https://alphacephei.com/vosk/models/vosk-model-cn-kaldi-multicn-0.15.zip
  

  

• 修改测试代码

  vi /opt/vosk/vosk-api/python/example/test_simple.py
  

  注释掉如下代码：

  # rec.SetPartialWords(True)
  

  要不会报如下错误：

  AttributeError: 'KaldiRecognizer' object has no attribute 'SetPartialWords'
  

• 配置模型
  解压 vosk-model-cn-0.15.zip 文件，并将解压后的文件夹名称修改为 model，随便命名也行，在test_simple.py文件内写入该地址model = Model("model")，最好写绝对路径，避免开机启动时找不到文件。

• 测试语音识别

  python3 test_simple.py test2.wav
  

• 使用Flask搭建简单识别接口

  import random
  import time
  from flask import Flask,request,json
  
  # vosk
  from vosk import Model, KaldiRecognizer, SetLogLevel
  import wave
  import os
  
  # system
  import sys
  import os
  import datetime
  import time
  import json
  import requests
  
  # 语音识别模型
  # print(os.path.dirname(__file__) + os.sep + "model")
  voskmodel = Model(os.path.dirname(__file__) + os.sep + "model")
  
  # 实例化一个web服务对象
  app = Flask(__name__)
  
  # 构造一个接受post请求的响应
  @app.route('/',methods=['POST'])
  def postRequest():
  	data = {}
  	data['data'] = {'text':'Not Found!'}
  	data['code'] = 404
  	path = request.form.get('path')
  	typ = request.form.get('typ') # url/path
  
  	if typ == 'url':
  		downlaod(path, os.path.dirname(__file__) + os.sep + 'voice.wav')
  		res = getwav(voskmodel, os.path.dirname(__file__) + os.sep + 'voice.wav')
  		data['data'] = res
  		data['code'] = 200
  	elif typ == 'path':
  		res = getwav(voskmodel, path)
  		data['data'] = res
  		data['code'] = 200
  	return data
  
  def downlaod(url, file_path):
    headers = {
  	"User-Agent": "Mozilla/5.0 (Windows NT 10.0; WOW64; rv:68.0) Gecko/20100101 Firefox/68.0"
    }
    r = requests.get(url=url, headers=headers)
    with open(file_path, "wb") as f:
  	f.write(r.content)
  	f.flush()
  
  def getwav(model,path):
  	startTime = datetime.datetime.now()
  	print('识别开始：'+ startTime.strftime('%Y-%m-%d %H:%M:%S'))
  	SetLogLevel(0)
  	wf = wave.open(path, "rb")
  	if wf.getnchannels() != 1 or wf.getsampwidth() != 2 or wf.getcomptype() != "NONE":
  		print ("Audio file must be WAV format mono PCM.")
  		exit (1)
  	rec = KaldiRecognizer(model, wf.getframerate())
  	rec.SetWords(True)
  
  	while True:
  		data = wf.readframes(4000)
  		if len(data) == 0:
  			break
  		if rec.AcceptWaveform(data):
  			rec.Result()
  		else:
  			rec.PartialResult()
  	res = json.loads(rec.FinalResult())
  	print('识别结果：'+ str(res['text']))
  	endTime = datetime.datetime.now()
  	print('识别结束：'+ endTime.strftime('%Y-%m-%d %H:%M:%S'))
  	useTime = str(int((endTime-startTime).total_seconds()))
  	print('识别耗时：' + useTime + '秒')
  	res['time'] = useTime
  	return res
  
  if __name__ == '__main__':
  	# 运行服务，并确定服务运行的IP和端口
  	app.run('127.0.0.1', '9090')
  

两种识别模式：

• 远程识别，通过拉取音频文件识别

{
	'path': 'http://127.0.0.1:8000/files/2022/08/xpxpgbcl.wav'
	'typ' : 'url'
}

• 本地识别，通过获取本地路径识别

{
	'path': 'E:\……/xpxpgbcl.wav'
	'typ' : 'path'
}