Baidu API call (three)-speech recognition

1. Open Baidu voice technology interface service

Basic process:

1. Open Baidu ai open platform https://ai.baidu.com/

2. Open the console

3. Choose voice technology

4. Choose to create an application.

5. Select the required service and simply fill in the application description to create it.

6. After the creation is complete, you can manage or delete the application you created in the application list.

7. You need to open the corresponding service before you can use it and


click to receive it immediately. The number of free calls can be obtained (hundreds of thousands of times are enough for personal learning and use)

Second, python realizes Baidu speech recognition

Example: (Send the saved or currently recorded voice file and then voice recognition)

1. Realize the function

Record for five seconds, then save the recording to the computer and send the speech recognition request to Baidu speech recognition server, get the speech recognition result and print it out

2. Code (commented)

#！/usr/bin/env python
import requests
import json
import base64
import pyaudio
import wave
import os
import psutil


#首先配置必要的信息
def bat(voice_path):
    baidu_server = 'https://aip.baidubce.com/oauth/2.0/token?'  #获取token的server
    grant_type = 'client_credentials'
    client_id = '' #API KEY
    client_secret = '' #Secret KEY   这里可以自己去百度注册，这里是我的API KEY 和 Secret KEY

    #合成请求token的url
    url = baidu_server+'grant_type='+grant_type+'&client_id='+client_id+'&client_secret='+client_secret

    #获取token
    res = requests.get(url).text
    data = json.loads(res)  #将json格式转换为字典格式
    token = data['access_token']

    #设置音频的属性，采样率，格式等
    VOICE_RATE = 8000
    FILE_NAME = voice_path
    # USER_ID = '16241950' #这里的id随便填填就好啦，我填的自己昵称
    FILE_TYPE = 'wav'
    CUID="wate_play"   #用户唯一标识符，用来区分用户，可以修改
    #读取文件二进制内容
    f_obj = open(FILE_NAME, 'rb')
    content = base64.b64encode(f_obj.read())   # 百度语音识别需要base64编码格式
    speech = content.decode("utf-8")
    size = os.path.getsize(FILE_NAME)

    #json封装
    datas = json.dumps({
    
          #json.dumps将一个Python数据结构转换为JSON ； json.loads将一个JSON编码的字符串转换回一个Python数据结构 
        'format': FILE_TYPE,
        'rate': VOICE_RATE,
        'channel': 1,
        'cuid': CUID,
        'token': token,   #上面从百度平台获取的token信息
        'speech': speech,
        'len': size,
        "dev_pid":"1537"
    })
    return datas

#设置headers和请求地址url
def post(datas):
    headers = {
    
    'Content-Type':'application/json'}
    #url = 'https://aip.baidubce.com/oauth/2.0/token?'  #技术文档中这个是获取token的url
    url = "http://vop.baidu.com/server_api"   #技术文档中给出的这个是语音识别的服务器接口

    #用post方法传数据
    request = requests.post(url, datas, headers)
    result = json.loads(request.text)
    print('result:',result)
    text = result.get("result")
    if result['err_no'] == 0:
        return text
    else:
        return "Error"

#录音并将录音结果保存到filepath处in_path = C:\voice\voice.wav
def get_audio(filepath):
    input("回车开始录音 >>>")     #输出提示文本，input接收一个值,转为str，赋值给aa
    CHUNK = 256                 #定义数据流块(每个数据块儿存放位数，正好为一个字节)
    FORMAT = pyaudio.paInt16    #量化位数（音量级划分）
    CHANNELS = 1               # 声道数;声道数：可以是单声道或者是双声道
    RATE = 8000                # 采样率;采样率：一秒内对声音信号的采集次数，常用的有8kHz, 16kHz, 32kHz, 48kHz, 11.025kHz, 22.05kHz, 44.1kHz
    RECORD_SECONDS = 5          #录音秒数
    WAVE_OUTPUT_FILENAME = filepath     #wav文件路径
    p = pyaudio.PyAudio()               #实例化

    stream = p.open(format=FORMAT,
                    channels=CHANNELS,
                    rate=RATE,
                    input=True,
                    frames_per_buffer=CHUNK)
    print("*"*10, "开始录音：请在5秒内输入语音")
    frames = []                                                 #定义一个列表
    for i in range(0, int(RATE / CHUNK * RECORD_SECONDS)):      #循环，采样率(8000 / 256) * 5 = (8000*5) / 256
        data = stream.read(CHUNK)                               #读取chunk个字节 保存到data中
        frames.append(data)                                     #向列表frames中添加数据data
    # print(frames)
    print("*" * 10, "录音结束\n")

    stream.stop_stream()
    stream.close()          #关闭
    p.terminate()           #终结

    wf = wave.open(WAVE_OUTPUT_FILENAME, 'wb')                  #打开wav文件创建一个音频对象wf，开始写WAV文件
    wf.setnchannels(CHANNELS)                                   #配置声道数
    wf.setsampwidth(p.get_sample_size(FORMAT))                  #配置量化位数
    wf.setframerate(RATE)                                       #配置采样率
    wf.writeframes(b''.join(frames))                            #转换为二进制数据写入文件
    wf.close()              #关闭
    return

def check_disk():
    list_drive = psutil.disk_partitions()  # 找出本地磁盘列表，保存的是结构体对象
    list_disk = []
    for drive in list_drive:
        list_disk.append(drive.device)
    return list_disk

if __name__ == '__main__':
    list_disk = check_disk() # 检索本地磁盘
    dirname_path = os.path.join(list_disk[0], "voice") # 设置语音文件存放路径
    
    if not os.path.exists(dirname_path):# 如果不存在该文件就创建一个该文件夹
        os.makedirs(dirname_path)

    filename = "voice.wav"  # 定义语音文件名
    in_path = os.path.join(dirname_path, filename)  # 得到路径为：in_path = C:\voice\voice.wav
    #print('in_path:',in_path)   #录音文件保存在in_path: C:\voice\voice.wav
    get_audio(in_path) # 录音

    datas = bat(in_path) # 封装百度语音识别需要的配置信息，返回请求头
    res = post(datas) # 连接百度语音识别接口，得到识别结果
    #print("识别结果：",res)
    print("识别结果：",res[0])

At last

For the calls of other Baidu API series, please refer to https://blog.csdn.net/weixin_45386875/article/details/113705329