[深度学习] 百度api的asr语音识别(包括长音频，识别多个语句)

以下代码包括了ocr , asr短音频，asr长音频（多个语句）。
正常调用asr的话，效果不好，所以用asr长音频（多个语句切分调用），效果可观很多。

import aip
from aip import ocr
from aip import speech
import os

import wave                        #音频文件处理
import urllib.request, pycurl
#import base64  
import json
# get access token by api key & secret key  
import time
import re

ocr

# https://console.bce.baidu.com/ai 百度智能云 创建ocr的api
ocr_APP_ID = 'XXXXXXX' 
ocr_API_KEY = 'XXXXXXXXXXXXXXXXXXXX'
ocr_SECRET_KEY = 'XXXXXXXXXXXXXXXXXXXXXXXXXXXX'
ocr_client = ocr.AipOcr(APP_ID, API_KEY, SECRET_KEY)

def get_file_content(filePath):
    with open(filePath, 'rb') as fp:
        return fp.read()

def image2text(fileName):
    image = get_file_content(fileName)
    dic_result = ocr_client.basicGeneral(image)
    try:
        res = dic_result['words_result']
    except:
        res = []
    result = ''
    for m in res:
        result = result + str(m['words'])
    return result

asr(短视频)

# https://console.bce.baidu.com/ai 百度智能云 创建asr的api
asr_APP_ID = 'XXXXXXXXXXX' 
asr_API_KEY = "XXXXXXXXXXXXXXXXXXXXXXXXXX"           
asr_SECRET_KEY = "XXXXXXXXXXXXXXXXXXXXXXXXXXXX" 
asr_client = speech.AipSpeech(asr_APP_ID, asr_API_KEY, asr_SECRET_KEY)

# def wav2text(fileName):
#     fp = wave.open(fileName, 'rb') 
#     nf = fp.getnframes()
#     # f_len = nf * 2    
#     wav = fp.readframes(nf) #audio_data
#     dic_result = asr_client.asr(wav)
# #     try:
# #         res = dic_result['words_result']
# #     except:
# #         res = []
# #     result = ''
# #     for m in res:
# #         result = result + str(m['words'])
#     return dic_result

# 5.短时间举例子
# 语音参数 必须符合16k或8K采样率、16bit采样位数、单声道
# 语音格式 PCM、WAV、AMR
def baidu_Speech_To_Text(filePath):  # 百度语音识别
    asr_APP_ID = 'XXXXXXXX' 
    asr_API_KEY = "XXXXXXXXXXXXXXXXXXXXXX"            #这两行是登录用的密码
    asr_SECRET_KEY = "XXXXXXXXXXXXXXXXXXXXXXXXXX" 
    asr_client = speech.AipSpeech(asr_APP_ID, asr_API_KEY, asr_SECRET_KEY)
    # 读取文件
    with open(filePath, 'rb') as fp:
        audioPcm = fp.read()
    json = asr_client.asr(audioPcm, 'wav', 16000, {'lan': 'zh', })
    print(json)
    if 'success' in json['err_msg']:
        context = json['result'][0]
        print('成功，返回结果为：', context)
    else:
        context = '=====识别失败====='
        print('识别失败！')
    return context

asr(长视频（其实是多语句）)

from pydub import AudioSegment
from pydub.utils import mediainfo
from aip import speech

def baidu_Speech_To_Text(filePath):  # 百度语音识别
    asr_APP_ID = 'XXXXXXXX' 
    asr_API_KEY = "XXXXXXXXXXXXXXXXXXXXXX"            #这两行是登录用的密码
    asr_SECRET_KEY = "XXXXXXXXXXXXXXXXXXXXXXXXXX" 
    asr_client = speech.AipSpeech(asr_APP_ID, asr_API_KEY, asr_SECRET_KEY)
    # 读取文件
    with open(filePath, 'rb') as fp:
        audioPcm = fp.read()
    json = asr_client.asr(audioPcm, 'wav', 16000, {'lan': 'zh', })
    print(json)
    if 'success' in json['err_msg']:
        context = json['result'][0]
        print('成功，返回结果为：', context)
    else:
        context = '=====识别失败====='
        print('识别失败！')
    return context
def sound_cut(file_name):
    if os.path.exists('识别结果.txt'):
        os.remove(r'识别结果.txt')
    song = mediainfo(file_name)
    song_length = str(int(float(song['duration'])))  # 读取文件时长
    song_size = str(round(float(int(song['size']) / 1024 / 1024), 2)) + 'M'  # 读取文件大小保留两位小数round(变量,2)
    song_filename = song['filename']  # 读取文件地址
    song_format_name = song['format_name']  # 读取文件格式
    print('\t长度', song_length, '\t文件大小', song_size, '\t文件路径', song_filename, '\t文件格式', song_format_name)
    cut_song_num = int(int(song_length) / 59) + 1  # 每段59s，计算切割段数
    print('切割次数', cut_song_num)
    sound = AudioSegment.from_mp3(file_name)
    # 单位：ms
    stat_time = 0
    end_time = 59
    for i in range(cut_song_num):
        if i == cut_song_num - 1:  # 判断如果是最后一次截断
            cut_song = sound[stat_time * 1000:]  # 截取到最后的时间
            end_time = int(song_length)
        else:
            cut_song = sound[stat_time * 1000:end_time * 1000]
        save_name = r"temp-" + str(i + 1) + '.mp3'  # 设置文件保存名称
        cut_song.export(save_name, format="mp3")  # 进行切割
        save_name_pcm = r"temp-" + str(i + 1) + '.wav'  # 设置文件保存名称
        mp3_version = AudioSegment.from_mp3(save_name)  # 可以根据文件不太类型导入不同from方法
        mono = mp3_version.set_frame_rate(16000).set_channels(1)  # 设置声道和采样率
        mono.export(save_name_pcm, format='wav', codec='pcm_s16le')  # codec此参数本意是设定16bits pcm编码器, 但发现此参数可以省略
        context = baidu_Speech_To_Text(save_name_pcm)
        with open(r'识别结果.txt', 'a', encoding='utf-8') as f:
            f.write(context)
        os.remove(save_name)  # 删除mp3文件
        os.remove(save_name_pcm)  # 删除mp3文件
        print(save_name, 'end_time=', stat_time, 'end_time=', end_time)
        # 切割完加入下一段的参数
        stat_time += 59
        end_time += 59

参考： https://blog.csdn.net/qq_40584593/article/details/110311540