Python直接调用msc.dll完成语音识别asr

RYHAN

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

1. #! /usr/bin/env python
   
2. # coding=utf-8
   
3. 
   
4. __author__ = 'ryhan'
   
5. 
   
6. 
   
7. # 以下代码解决输出乱码问题
   
8. import sys
   
9. # print sys.getdefaultencoding()
   
10. reload(sys)
    
11. sys.setdefaultencoding('utf-8')
    
12. # print sys.getdefaultencoding()
    
13. 
    
14. from ctypes import *
    
15. import time
    
16. 
    
17. sys.path.append('.')
    
18. 
    
19. dll = windll.LoadLibrary("msc.dll")
    
20. 
    
21. # print dll.MSPLogin
    
22. # print dll.QTTSSessionBegin
    
23. # print dll.QTTSTextPut
    
24. # print dll.QTTSAudioGet
    
25. # print dll.QTTSSessionEnd
    
26. 
    
27. uname = "973214924@qq.com"
    
28. upass = ""
    
29. 
    
30. login_params = "appid = 自己的id, work_dir = ."
    
31. 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"
    
32. grammar_id = None  # '346c55176c3a5fc69750a3068b1a8457'
    
33. 
    
34. FRAME_LEN = 640  # Byte
    
35. 
    
36. MSP_SUCCESS = 0
    
37. # 端点数据
    
38. MSP_EP_LOOKING_FOR_SPEECH = 0
    
39. MSP_EP_IN_SPEECH = 1
    
40. MSP_EP_AFTER_SPEECH = 3
    
41. MSP_EP_TIMEOUT = 4
    
42. MSP_EP_ERROR = 5
    
43. MSP_EP_MAX_SPEECH = 6
    
44. MSP_EP_IDLE = 7
    
45. # 音频情况
    
46. MSP_AUDIO_SAMPLE_INIT = 0x00
    
47. MSP_AUDIO_SAMPLE_FIRST = 0x01
    
48. MSP_AUDIO_SAMPLE_CONTINUE = 0x02
    
49. MSP_AUDIO_SAMPLE_LAST = 0x04
    
50. # 识别状态
    
51. MSP_REC_STATUS_SUCCESS = 0
    
52. MSP_REC_STATUS_NO_MATCH = 1
    
53. MSP_REC_STATUS_INCOMPLETE = 2
    
54. MSP_REC_STATUS_NON_SPEECH_DETECTED = 3
    
55. MSP_REC_STATUS_SPEECH_DETECTED = 4
    
56. MSP_REC_STATUS_COMPLETE = 5
    
57. MSP_REC_STATUS_MAX_CPU_TIME = 6
    
58. MSP_REC_STATUS_MAX_SPEECH = 7
    
59. MSP_REC_STATUS_STOPPED = 8
    
60. MSP_REC_STATUS_REJECTED = 9
    
61. MSP_REC_STATUS_NO_SPEECH_FOUND = 10
    
62. MSP_REC_STATUS_FAILURE = MSP_REC_STATUS_NO_MATCH
    
63. 
    
64. filename = "tts_sample.wav"
    
65. filename = "iflytek01.wav"
    
66. # filename = "ryhan.wav"
    
67. 
    
68. 
    
69. class Msp:
    
70.     def __init__(self):
    
71.         pass
    
72. 
    
73.     def login(self):
    
74.         ret = dll.MSPLogin(None, None, login_params)
    
75.         print('MSPLogin =>'), ret
    
76. 
    
77.         pass
    
78. 
    
79.     def isr(self, audiofile, session_begin_params, grammar_id):
    
80.         ret = c_int()
    
81.         sessionID = dll.QISRSessionBegin(grammar_id, session_begin_params, byref(ret))
    
82.         print 'QISRSessionBegin => sessionID:', sessionID, 'ret:', ret.value
    
83. 
    
84.         # 每秒【1000ms】  16000 次 * 16 bit 【20B】 ，每毫秒：1.6 * 16bit 【1.6*2B】 = 32Byte
    
85.         # 1帧音频20ms【640B】 每次写入 10帧=200ms 【6400B】
    
86. 
    
87.         piceLne = FRAME_LEN * 20
    
88.         epStatus = c_int(0)
    
89.         recogStatus = c_int(0)
    
90. 
    
91.         wavFile = open(audiofile, 'rb')
    
92.         wavData = wavFile.read(piceLne)
    
93. 
    
94.         ret = dll.QISRAudioWrite(sessionID, wavData, len(wavData), MSP_AUDIO_SAMPLE_FIRST, byref(epStatus),
    
95.                                  byref(recogStatus))
    
96.         print 'len(wavData):', len(
    
97.             wavData), 'QISRAudioWrite ret:', ret, 'epStatus:', epStatus, 'recogStatus:', recogStatus
    
98. 
    
99.         while wavData:
    
100.             wavData = wavFile.read(piceLne)
     
101. 
     
102.             if len(wavData) == 0:
     
103.                 break
     
104. 
     
105.             ret = dll.QISRAudioWrite(sessionID, wavData, len(wavData), MSP_AUDIO_SAMPLE_CONTINUE, byref(epStatus),
     
106.                                      byref(recogStatus))
     
107.             print 'len(wavData):', len(
     
108.                 wavData), 'QISRAudioWrite ret:', ret, 'epStatus:', epStatus, 'recogStatus:', recogStatus
     
109.             time.sleep(0.2)
     
110. 
     
111.         ret = dll.QISRAudioWrite(sessionID, None, 0, MSP_AUDIO_SAMPLE_LAST, byref(epStatus),
     
112.                                  byref(recogStatus))
     
113.         print 'len(wavData):', len(
     
114.             wavData), 'QISRAudioWrite ret:', ret, 'epStatus:', epStatus, 'recogStatus:', recogStatus
     
115. 
     
116. 
     
117.         # -- 获取音频
     
118.         laststr = ''
     
119.         while recogStatus.value != MSP_REC_STATUS_COMPLETE:
     
120.             ret = c_int()
     
121.             dll.QISRGetResult.restype = c_char_p
     
122.             retstr = dll.QISRGetResult(sessionID, byref(recogStatus), 0, byref(ret))
     
123.             if retstr is not None:
     
124.                 laststr += retstr
     
125.             print 'ret:', ret, 'recogStatus:', recogStatus
     
126. 
     
127.             time.sleep(0.2)
     
128. 
     
129.         print '*' * 20, 'laststr:', '*' * 20
     
130.         print laststr
     
131.         print '*' * 20, 'laststr:', '*' * 20
     
132. 
     
133.     pass
     
134. 
     
135. 
     
136. if __name__ == "__main__":
     
137.     msp = Msp()
     
138. 
     
139.     msp.login()
     
140.     msp.isr(filename, session_begin_params, grammar_id)

复制代码

HandlerMscIsr_v2.py

1. #! /usr/bin/env python
   
2. # coding=utf-8
   
3. 
   
4. __author__ = 'ryhan'
   
5. 
   
6. 
   
7. # 以下代码解决输出乱码问题
   
8. import sys
   
9. # print sys.getdefaultencoding()
   
10. reload(sys)
    
11. sys.setdefaultencoding('utf-8')
    
12. # print sys.getdefaultencoding()
    
13. 
    
14. from ctypes import *
    
15. import time
    
16. import pyaudio
    
17. 
    
18. sys.path.append('.')
    
19. 
    
20. dll = windll.LoadLibrary("msc.dll")
    
21. 
    
22. # print dll.MSPLogin
    
23. # print dll.QTTSSessionBegin
    
24. # print dll.QTTSTextPut
    
25. # print dll.QTTSAudioGet
    
26. # print dll.QTTSSessionEnd
    
27. 
    
28. uname = "973214924@qq.com"
    
29. upass = ""
    
30. 
    
31. login_params = "appid = 自己的id, work_dir = ."
    
32. session_begin_params = "sub=iat,aue=speex-wb;7,result_type=plain,result_encoding=utf8,language=zh_cn," \
    
33.                        "accent=mandarin,sample_rate=16000,domain=music,vad_bos=2000,vad_eos=2000"
    
34. grammar_id = None  # '346c55176c3a5fc69750a3068b1a8457'
    
35. 
    
36. FRAME_LEN = 640  # Byte
    
37. 
    
38. MSP_SUCCESS = 0
    
39. # 端点数据
    
40. MSP_EP_LOOKING_FOR_SPEECH = 0
    
41. MSP_EP_IN_SPEECH = 1
    
42. MSP_EP_AFTER_SPEECH = 3
    
43. MSP_EP_TIMEOUT = 4
    
44. MSP_EP_ERROR = 5
    
45. MSP_EP_MAX_SPEECH = 6
    
46. MSP_EP_IDLE = 7
    
47. # 音频情况
    
48. MSP_AUDIO_SAMPLE_INIT = 0x00
    
49. MSP_AUDIO_SAMPLE_FIRST = 0x01
    
50. MSP_AUDIO_SAMPLE_CONTINUE = 0x02
    
51. MSP_AUDIO_SAMPLE_LAST = 0x04
    
52. # 识别状态
    
53. MSP_REC_STATUS_SUCCESS = 0
    
54. MSP_REC_STATUS_NO_MATCH = 1
    
55. MSP_REC_STATUS_INCOMPLETE = 2
    
56. MSP_REC_STATUS_NON_SPEECH_DETECTED = 3
    
57. MSP_REC_STATUS_SPEECH_DETECTED = 4
    
58. MSP_REC_STATUS_COMPLETE = 5
    
59. MSP_REC_STATUS_MAX_CPU_TIME = 6
    
60. MSP_REC_STATUS_MAX_SPEECH = 7
    
61. MSP_REC_STATUS_STOPPED = 8
    
62. MSP_REC_STATUS_REJECTED = 9
    
63. MSP_REC_STATUS_NO_SPEECH_FOUND = 10
    
64. MSP_REC_STATUS_FAILURE = MSP_REC_STATUS_NO_MATCH
    
65. 
    
66. filename = "tts_sample.wav"
    
67. filename = "iflytek01.wav"
    
68. # filename = "ryhan.wav"
    
69. 
    
70. 
    
71. class Msp:
    
72.     def __init__(self):
    
73.         pass
    
74. 
    
75.     def login(self):
    
76.         ret = dll.MSPLogin(None, None, login_params)
    
77.         print('MSPLogin =>'), ret
    
78. 
    
79.         pass
    
80. 
    
81.     def isr(self, audiofile, session_begin_params, grammar_id):
    
82.         ret = c_int()
    
83.         sessionID = dll.QISRSessionBegin(grammar_id, session_begin_params, byref(ret))
    
84.         print 'QISRSessionBegin => sessionID:', sessionID, 'ret:', ret.value
    
85. 
    
86.         # 每秒【1000ms】  16000 次 * 16 bit 【20B】 ，每毫秒：1.6 * 16bit 【1.6*2B】 = 32Byte
    
87.         # 1帧音频20ms【640B】 每次写入 10帧=200ms 【6400B】
    
88. 
    
89. 
    
90.         p = pyaudio.PyAudio()
    
91.         stream = p.open(rate=16000,
    
92.                         channels=1,
    
93.                         format=pyaudio.paInt16,
    
94.                         input=True,
    
95.                         frames_per_buffer=2048)
    
96. 
    
97.         epStatus = c_int(0)
    
98.         recogStatus = c_int(0)
    
99. 
    
100.         wavData = stream.read(2048)
     
101. 
     
102.         ret = dll.QISRAudioWrite(sessionID, wavData, len(wavData), MSP_AUDIO_SAMPLE_FIRST, byref(epStatus),
     
103.                                  byref(recogStatus))
     
104.         print 'len(wavData):', len(
     
105.             wavData), 'QISRAudioWrite ret:', ret, 'epStatus:', epStatus, 'recogStatus:', recogStatus
     
106. 
     
107.         times = 0
     
108.         while True:
     
109.             wavData = stream.read(2048)
     
110.             ret = dll.QISRAudioWrite(sessionID, wavData, len(wavData), MSP_AUDIO_SAMPLE_CONTINUE, byref(epStatus),
     
111.                                      byref(recogStatus))
     
112.             print 'len(wavData):', len(
     
113.                 wavData), 'QISRAudioWrite ret:', ret, 'epStatus:', epStatus, 'recogStatus:', recogStatus
     
114.             time.sleep(0.02)
     
115.             times += 1
     
116. 
     
117.             if epStatus.value != MSP_EP_IN_SPEECH:
     
118.                 break
     
119. 
     
120.         ret = dll.QISRAudioWrite(sessionID, None, 0, MSP_AUDIO_SAMPLE_LAST, byref(epStatus),
     
121.                                  byref(recogStatus))
     
122.         print 'len(wavData):', len(
     
123.             wavData), 'QISRAudioWrite ret:', ret, 'epStatus:', epStatus, 'recogStatus:', recogStatus
     
124. 
     
125. 
     
126.         # -- 获取音频
     
127.         laststr = ''
     
128.         while recogStatus.value != MSP_REC_STATUS_COMPLETE:
     
129.             ret = c_int()
     
130.             dll.QISRGetResult.restype = c_char_p
     
131.             retstr = dll.QISRGetResult(sessionID, byref(recogStatus), 0, byref(ret))
     
132.             if retstr is not None:
     
133.                 laststr += retstr
     
134.             print 'ret:', ret, 'recogStatus:', recogStatus
     
135. 
     
136.             time.sleep(0.2)
     
137. 
     
138.         print '*' * 20, 'laststr:', '*' * 20
     
139.         print laststr
     
140.         print '*' * 20, 'laststr:', '*' * 20
     
141. 
     
142.     pass
     
143. 
     
144. 
     
145. if __name__ == "__main__":
     
146.     msp = Msp()
     
147. 
     
148.     msp.login()
     
149.     msp.isr(filename, session_begin_params, grammar_id)

复制代码

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

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

msc.dll也在附件里吧？是不是C++的动态链接库啊

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........

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。

不错不错，学习了

厉害厉害，正找这个，学习一下

楼主厉害，真需要这个，果断支持

正好用上，谢谢楼主分享