Python調用百度語音識別REST API

nxhujiee 2016-03-27

展開全文

百度語音識別通過 REST API 的方式給開發(fā)者提供一個通用的 HTTP 接口，基于該接口，開發(fā)者可以輕松的獲取語音識別能力。SDK中只提供了PHP、C和JAVA的相關樣例，然而個人以為，使用Python開發(fā)難度更低，本文描述了簡單使用Python調用百度語音識別服務 REST API 的簡單樣例。

注冊開發(fā)者賬號和創(chuàng)建應用不再贅述，百度的REST API在調用過程基本分為三步：

獲取token
提交數(shù)據(jù)
處理JSON

獲取token

#Api_Key申請的api_key
#Secrect_Key對應的secret_key
url = 'https://openapi.baidu.com/oauth/2.0/token?grant_type=client_credentials&client_id='+Api_Key+'&client_secret='+Secrect_Key
res = urllib2.urlopen(url).read()
data = json.loads(res)
token = data['access_token']    #獲取的token
print 'token獲取成功：'+`token`

提交數(shù)據(jù)

VOICE_RATE = 8000
WAVE_FILE = 'temp.wav'
USER_ID = 'duvoice'
WAVE_TYPE = 'wav'
#其它參數(shù)可參考sdk文檔

f = open(WAVE_FILE,'r')
speech = base64.b64encode(f.read())
size = os.path.getsize(WAVE_FILE)
update = json.dumps({'format':WAVE_TYPE,'rate':VOICE_RATE,'channel':1,'cuid':USER_ID,'token':token,'speech':speech,'len':size})
r = urllib2.urlopen(self.url,update)

處理JSON

t = r.read()
result = json.loads(t)
if result['err_msg']=='success.':
    word = result['result'][0].encode('utf-8')
    if word!='':
        if word[len(word)-3:len(word)]=='，':
            print word[0:len(word)-3]
            return word[0:len(word)-3]
        else:
            print word
            return word
    else:
        print "音頻文件不存在或格式錯誤"
        return ''
else:
    print "錯誤"

經(jīng)過以上步驟便可提交一個wav文件進行解析，同時也可使用python的pyaudio庫進行錄音：

#!/usr/bin/env python
# -*- coding: utf-8 -*-

from pyaudio import PyAudio, paInt16 
import numpy as np 
from datetime import datetime 
import wave

class recoder:
    NUM_SAMPLES = 2000      #pyaudio內置緩沖大小
    SAMPLING_RATE = 8000    #取樣頻率
    LEVEL = 500         #聲音保存的閾值
    COUNT_NUM = 20      #NUM_SAMPLES個取樣之內出現(xiàn)COUNT_NUM個大于LEVEL的取樣則記錄聲音
    SAVE_LENGTH = 8         #聲音記錄的最小長度：SAVE_LENGTH * NUM_SAMPLES 個取樣
    TIME_COUNT = 60     #錄音時間，單位s

    Voice_String = []

    def savewav(self,filename):
        wf = wave.open(filename, 'wb') 
        wf.setnchannels(1) 
        wf.setsampwidth(2) 
        wf.setframerate(self.SAMPLING_RATE) 
        wf.writeframes("".join(self.Voice_String)) 
        wf.close() 

    def recoder(self):
        pa = PyAudio() 
        stream = pa.open(format=paInt16, channels=1, rate=self.SAMPLING_RATE, input=True, 
            frames_per_buffer=self.NUM_SAMPLES) 
        save_count = 0 
        save_buffer = [] 
        time_count = self.TIME_COUNT

        while True:
            time_count -= 1
            # print time_count
            # 讀入NUM_SAMPLES個取樣
            string_audio_data = stream.read(self.NUM_SAMPLES) 
            # 將讀入的數(shù)據(jù)轉換為數(shù)組
            audio_data = np.fromstring(string_audio_data, dtype=np.short)
            # 計算大于LEVEL的取樣的個數(shù)
            large_sample_count = np.sum( audio_data > self.LEVEL )
            print np.max(audio_data)
            # 如果個數(shù)大于COUNT_NUM，則至少保存SAVE_LENGTH個塊
            if large_sample_count > self.COUNT_NUM:
                save_count = self.SAVE_LENGTH 
            else: 
                save_count -= 1

            if save_count < 0:
                save_count = 0 

            if save_count > 0 : 
            # 將要保存的數(shù)據(jù)存放到save_buffer中
                #print  save_count > 0 and time_count >0
                save_buffer.append( string_audio_data ) 
            else: 
            #print save_buffer
            # 將save_buffer中的數(shù)據(jù)寫入WAV文件，WAV文件的文件名是保存的時刻
                #print "debug"
                if len(save_buffer) > 0 : 
                    self.Voice_String = save_buffer
                    save_buffer = [] 
                    print "Recode a piece of  voice successfully!"
                    return True
            if time_count==0: 
                if len(save_buffer)>0:
                    self.Voice_String = save_buffer
                    save_buffer = [] 
                    print "Recode a piece of  voice successfully!"
                    return True
                else:
                    return False