Python直接调用msc.dll完成语音识别asr

RYHAN · RYHAN 发表于 2016-11-11 17:40:16

本帖最后由 RYHAN 于 2018-5-3 20:58 编辑

在上次直接调用dll实现tts后，持续完成了生成音频以及直接调用耳机播放合成的音频。
今天抽空研究对接asr，代码一共两个版本：

v1.读取音频，调用msc
v2.读取耳麦，调用msc

ps：过程中参考论坛内其他同学的代码，并完成demo编写

HandlerMscIsr.py

#! /usr/bin/env python
# coding=utf-8
__author__ = 'ryhan'
# 以下代码解决输出乱码问题
import sys
# print sys.getdefaultencoding()
reload(sys)
sys.setdefaultencoding('utf-8')
# print sys.getdefaultencoding()
from ctypes import *
import time
sys.path.append('.')
dll = windll.LoadLibrary("msc.dll")
# print dll.MSPLogin
# print dll.QTTSSessionBegin
# print dll.QTTSTextPut
# print dll.QTTSAudioGet
# print dll.QTTSSessionEnd
uname = "973214924@qq.com"
upass = ""
login_params = "appid = 自己的id, work_dir = ."
session_begin_params = "sub=iat,aue=speex-wb;7,result_type=plain,result_encoding=utf8,language=zh_cn,accent=mandarin,sample_rate=16000,domain=music,vad_bos=1000,vad_eos=1000"
grammar_id = None # '346c55176c3a5fc69750a3068b1a8457'
FRAME_LEN = 640 # Byte
MSP_SUCCESS = 0
# 端点数据
MSP_EP_LOOKING_FOR_SPEECH = 0
MSP_EP_IN_SPEECH = 1
MSP_EP_AFTER_SPEECH = 3
MSP_EP_TIMEOUT = 4
MSP_EP_ERROR = 5
MSP_EP_MAX_SPEECH = 6
MSP_EP_IDLE = 7
# 音频情况
MSP_AUDIO_SAMPLE_INIT = 0x00
MSP_AUDIO_SAMPLE_FIRST = 0x01
MSP_AUDIO_SAMPLE_CONTINUE = 0x02
MSP_AUDIO_SAMPLE_LAST = 0x04
# 识别状态
MSP_REC_STATUS_SUCCESS = 0
MSP_REC_STATUS_NO_MATCH = 1
MSP_REC_STATUS_INCOMPLETE = 2
MSP_REC_STATUS_NON_SPEECH_DETECTED = 3
MSP_REC_STATUS_SPEECH_DETECTED = 4
MSP_REC_STATUS_COMPLETE = 5
MSP_REC_STATUS_MAX_CPU_TIME = 6
MSP_REC_STATUS_MAX_SPEECH = 7
MSP_REC_STATUS_STOPPED = 8
MSP_REC_STATUS_REJECTED = 9
MSP_REC_STATUS_NO_SPEECH_FOUND = 10
MSP_REC_STATUS_FAILURE = MSP_REC_STATUS_NO_MATCH
filename = "tts_sample.wav"
filename = "iflytek01.wav"
# filename = "ryhan.wav"
class Msp:
def __init__(self):
pass
def login(self):
ret = dll.MSPLogin(None, None, login_params)
print('MSPLogin =>'), ret
pass
def isr(self, audiofile, session_begin_params, grammar_id):
ret = c_int()
sessionID = dll.QISRSessionBegin(grammar_id, session_begin_params, byref(ret))
print 'QISRSessionBegin => sessionID:', sessionID, 'ret:', ret.value
# 每秒【1000ms】 16000 次 * 16 bit 【20B】，每毫秒：1.6 * 16bit 【1.6*2B】 = 32Byte
# 1帧音频20ms【640B】每次写入 10帧=200ms 【6400B】
piceLne = FRAME_LEN * 20
epStatus = c_int(0)
recogStatus = c_int(0)
wavFile = open(audiofile, 'rb')
wavData = wavFile.read(piceLne)
ret = dll.QISRAudioWrite(sessionID, wavData, len(wavData), MSP_AUDIO_SAMPLE_FIRST, byref(epStatus),
byref(recogStatus))
print 'len(wavData):', len(
wavData), 'QISRAudioWrite ret:', ret, 'epStatus:', epStatus, 'recogStatus:', recogStatus
while wavData:
wavData = wavFile.read(piceLne)
if len(wavData) == 0:
break
ret = dll.QISRAudioWrite(sessionID, wavData, len(wavData), MSP_AUDIO_SAMPLE_CONTINUE, byref(epStatus),
byref(recogStatus))
print 'len(wavData):', len(
wavData), 'QISRAudioWrite ret:', ret, 'epStatus:', epStatus, 'recogStatus:', recogStatus
time.sleep(0.2)
ret = dll.QISRAudioWrite(sessionID, None, 0, MSP_AUDIO_SAMPLE_LAST, byref(epStatus),
byref(recogStatus))
print 'len(wavData):', len(
wavData), 'QISRAudioWrite ret:', ret, 'epStatus:', epStatus, 'recogStatus:', recogStatus
# -- 获取音频
laststr = ''
while recogStatus.value != MSP_REC_STATUS_COMPLETE:
ret = c_int()
dll.QISRGetResult.restype = c_char_p
retstr = dll.QISRGetResult(sessionID, byref(recogStatus), 0, byref(ret))
if retstr is not None:
laststr += retstr
print 'ret:', ret, 'recogStatus:', recogStatus
time.sleep(0.2)
print '*' * 20, 'laststr:', '*' * 20
print laststr
print '*' * 20, 'laststr:', '*' * 20
pass
if __name__ == "__main__":
msp = Msp()
msp.login()
msp.isr(filename, session_begin_params, grammar_id)

复制代码

HandlerMscIsr_v2.py

#! /usr/bin/env python
# coding=utf-8
__author__ = 'ryhan'
# 以下代码解决输出乱码问题
import sys
# print sys.getdefaultencoding()
reload(sys)
sys.setdefaultencoding('utf-8')
# print sys.getdefaultencoding()
from ctypes import *
import time
import pyaudio
sys.path.append('.')
dll = windll.LoadLibrary("msc.dll")
# print dll.MSPLogin
# print dll.QTTSSessionBegin
# print dll.QTTSTextPut
# print dll.QTTSAudioGet
# print dll.QTTSSessionEnd
uname = "973214924@qq.com"
upass = ""
login_params = "appid = 自己的id, work_dir = ."
session_begin_params = "sub=iat,aue=speex-wb;7,result_type=plain,result_encoding=utf8,language=zh_cn," \
"accent=mandarin,sample_rate=16000,domain=music,vad_bos=2000,vad_eos=2000"
grammar_id = None # '346c55176c3a5fc69750a3068b1a8457'
FRAME_LEN = 640 # Byte
MSP_SUCCESS = 0
# 端点数据
MSP_EP_LOOKING_FOR_SPEECH = 0
MSP_EP_IN_SPEECH = 1
MSP_EP_AFTER_SPEECH = 3
MSP_EP_TIMEOUT = 4
MSP_EP_ERROR = 5
MSP_EP_MAX_SPEECH = 6
MSP_EP_IDLE = 7
# 音频情况
MSP_AUDIO_SAMPLE_INIT = 0x00
MSP_AUDIO_SAMPLE_FIRST = 0x01
MSP_AUDIO_SAMPLE_CONTINUE = 0x02
MSP_AUDIO_SAMPLE_LAST = 0x04
# 识别状态
MSP_REC_STATUS_SUCCESS = 0
MSP_REC_STATUS_NO_MATCH = 1
MSP_REC_STATUS_INCOMPLETE = 2
MSP_REC_STATUS_NON_SPEECH_DETECTED = 3
MSP_REC_STATUS_SPEECH_DETECTED = 4
MSP_REC_STATUS_COMPLETE = 5
MSP_REC_STATUS_MAX_CPU_TIME = 6
MSP_REC_STATUS_MAX_SPEECH = 7
MSP_REC_STATUS_STOPPED = 8
MSP_REC_STATUS_REJECTED = 9
MSP_REC_STATUS_NO_SPEECH_FOUND = 10
MSP_REC_STATUS_FAILURE = MSP_REC_STATUS_NO_MATCH
filename = "tts_sample.wav"
filename = "iflytek01.wav"
# filename = "ryhan.wav"
class Msp:
def __init__(self):
pass
def login(self):
ret = dll.MSPLogin(None, None, login_params)
print('MSPLogin =>'), ret
pass
def isr(self, audiofile, session_begin_params, grammar_id):
ret = c_int()
sessionID = dll.QISRSessionBegin(grammar_id, session_begin_params, byref(ret))
print 'QISRSessionBegin => sessionID:', sessionID, 'ret:', ret.value
# 每秒【1000ms】 16000 次 * 16 bit 【20B】，每毫秒：1.6 * 16bit 【1.6*2B】 = 32Byte
# 1帧音频20ms【640B】每次写入 10帧=200ms 【6400B】
p = pyaudio.PyAudio()
stream = p.open(rate=16000,
channels=1,
format=pyaudio.paInt16,
input=True,
frames_per_buffer=2048)
epStatus = c_int(0)
recogStatus = c_int(0)
wavData = stream.read(2048)
ret = dll.QISRAudioWrite(sessionID, wavData, len(wavData), MSP_AUDIO_SAMPLE_FIRST, byref(epStatus),
byref(recogStatus))
print 'len(wavData):', len(
wavData), 'QISRAudioWrite ret:', ret, 'epStatus:', epStatus, 'recogStatus:', recogStatus
times = 0
while True:
wavData = stream.read(2048)
ret = dll.QISRAudioWrite(sessionID, wavData, len(wavData), MSP_AUDIO_SAMPLE_CONTINUE, byref(epStatus),
byref(recogStatus))
print 'len(wavData):', len(
wavData), 'QISRAudioWrite ret:', ret, 'epStatus:', epStatus, 'recogStatus:', recogStatus
time.sleep(0.02)
times += 1
if epStatus.value != MSP_EP_IN_SPEECH:
break
ret = dll.QISRAudioWrite(sessionID, None, 0, MSP_AUDIO_SAMPLE_LAST, byref(epStatus),
byref(recogStatus))
print 'len(wavData):', len(
wavData), 'QISRAudioWrite ret:', ret, 'epStatus:', epStatus, 'recogStatus:', recogStatus
# -- 获取音频
laststr = ''
while recogStatus.value != MSP_REC_STATUS_COMPLETE:
ret = c_int()
dll.QISRGetResult.restype = c_char_p
retstr = dll.QISRGetResult(sessionID, byref(recogStatus), 0, byref(ret))
if retstr is not None:
laststr += retstr
print 'ret:', ret, 'recogStatus:', recogStatus
time.sleep(0.2)
print '*' * 20, 'laststr:', '*' * 20
print laststr
print '*' * 20, 'laststr:', '*' * 20
pass
if __name__ == "__main__":
msp = Msp()
msp.login()
msp.isr(filename, session_begin_params, grammar_id)

复制代码

阿卜肚拉 · 阿卜肚拉发表于 2016-11-16 09:49:45

dll.QISRGetResult.restype = c_char_p
请问这句是做什么的呢，我如果注释掉的话，dll.QISRGetResult返回的就是null

retstr = dll.QISRGetResult(sessionID, byref(recogStatus), 0, byref(ret))
这里为什么每次返回的都是一串不同的数字呢

RYHAN · RYHAN 发表于 2016-11-19 16:36:17

阿卜肚拉发表于 2016-11-16 09:49
dll.QISRGetResult.restype = c_char_p
请问这句是做什么的呢，我如果注释掉的话，dll.QISRGetResult返回的 ...

dll.QISRGetResult.restype = c_char_p

这句话是定义获取的返回值类型，这个函数在c++里面定义的是char* 所以要这么用，让python自动包装返回值。

返回数字其实应该是内存的地址

lihuang1004@163 · lihuang1004@163 发表于 2016-11-24 21:38:34

msc.dll也在附件里吧？是不是C++的动态链接库啊

Waspss · Waspss 发表于 2016-11-25 22:12:53

阿卜肚拉发表于 2016-11-16 09:49
dll.QISRGetResult.restype = c_char_p
请问这句是做什么的呢，我如果注释掉的话，dll.QISRGetResult返回的 ...

厉害厉害........

Reidzhang · Reidzhang 发表于 2016-12-7 10:03:14

厉害厉害，感谢楼主分享，感觉最后都可以加个dll.QISRSessionEnd(sessionID,None)，这样重复调用不会报错。

1002466215@qq.c · 1002466215@qq.c 发表于 2017-3-8 09:44:23

不错不错，学习了

bigmaocxd · bigmaocxd 发表于 2017-4-22 10:44:56

厉害厉害，正找这个，学习一下

ljj1234 · ljj1234 发表于 2017-6-21 17:07:08

楼主厉害，真需要这个，果断支持

smartfrog · smartfrog 发表于 2017-6-28 22:30:43

正好用上，谢谢楼主分享

		自动登录	找回密码
密码			注册

小编推荐

推荐阅读

[demo分享] Python直接调用msc.dll完成语音识别asr

评分