Microsoft Speech to Text Python SDK SPXERR_INVALID_HEADER 问题

Question

我在使用 Microsoft Python Speech-to-Text Quickstart ("Quickstart: Recognize speech from an audio file") with the azure-cognitiveservices-speech v1.8.0 SDK 时遇到以下错误。

RuntimeError: Exception with an error code: 0xa (SPXERR_INVALID_HEADER)

此文件只有 3 个输入：

Azure 订阅密钥
Azure 服务区域
文件名

我正在使用以下测试 MP3 文件：

https://github.com/grokify/go-transcribe/blob/master/examples/mongodb-is-web-scale/web-scale_b2F-DItXtZs.mp3

这是完整的输出：

Traceback (most recent call last):
  File "main.py", line 16, in <module>
    speech_recognizer = speechsdk.SpeechRecognizer(speech_config=speech_config, audio_config=audio_input)
  File "/Library/Frameworks/Python.framework/Versions/3.7/lib/python3.7/site-packages/azure/cognitiveservices/speech/speech.py", line 761, in __init__
    self._impl = self._get_impl(impl.SpeechRecognizer, speech_config, audio_config)
  File "/Library/Frameworks/Python.framework/Versions/3.7/lib/python3.7/site-packages/azure/cognitiveservices/speech/speech.py", line 547, in _get_impl
    _impl = reco_type._from_config(speech_config._impl, audio_config._impl)
RuntimeError: Exception with an error code: 0xa (SPXERR_INVALID_HEADER)
[CALL STACK BEGIN]

3   libMicrosoft.CognitiveServices.Speech.core.dylib 0x0000000106ad88d2 CreateModuleObject + 1136482
4   libMicrosoft.CognitiveServices.Speech.core.dylib 0x0000000106ad7f4f CreateModuleObject + 1134047
5   libMicrosoft.CognitiveServices.Speech.core.dylib 0x00000001069d1803 CreateModuleObject + 59027
6   libMicrosoft.CognitiveServices.Speech.core.dylib 0x00000001069d1503 CreateModuleObject + 58259
7   libMicrosoft.CognitiveServices.Speech.core.dylib 0x0000000106a11c64 CreateModuleObject + 322292
8   libMicrosoft.CognitiveServices.Speech.core.dylib 0x0000000106a10be5 CreateModuleObject + 318069
9   libMicrosoft.CognitiveServices.Speech.core.dylib 0x0000000106a0e5a2 CreateModuleObject + 308274
10  libMicrosoft.CognitiveServices.Speech.core.dylib 0x0000000106a0e7c3 CreateModuleObject + 308819
11  libMicrosoft.CognitiveServices.Speech.core.dylib 0x0000000106960bc7 recognizer_create_speech_recognizer_from_config + 3863
12  libMicrosoft.CognitiveServices.Speech.core.dylib 0x000000010695fd74 recognizer_create_speech_recognizer_from_config + 196
13  _speech_py_impl.so                  0x00000001067ff35b PyInit__speech_py_impl + 814939
14  _speech_py_impl.so                  0x000000010679b530 PyInit__speech_py_impl + 405808
15  Python                              0x00000001060f65dc _PyMethodDef_RawFastCallKeywords + 668
16  Python                              0x00000001060f5a5a _PyCFunction_FastCallKeywords + 42
17  Python                              0x00000001061b45a4 call_function + 724
18  Python                              0x00000001061b1576 _PyEval_EvalFrameDefault + 25190
19  Python                              0x00000001060f5e90 function_code_fastcall + 128
20  Python                              0x00000001061b45b2 call_function + 738
21  Python                              0x00000001061b1576 _PyEval_EvalFrameDefault + 25190
22  Python                              0x00000001061b50d6 _PyEval_EvalCodeWithName + 2422
23  Python                              0x00000001060f55fb _PyFunction_FastCallDict + 523
24  Python                              0x00000001060f68cf _PyObject_Call_Prepend + 143
25  Python                              0x0000000106144d51 slot_tp_init + 145
26  Python                              0x00000001061406a9 type_call + 297
27  Python                              0x00000001060f5871 _PyObject_FastCallKeywords + 433
28  Python                              0x00000001061b4474 call_function + 420
29  Python                              0x00000001061b16bd _PyEval_EvalFrameDefault + 25517
30  Python                              0x00000001061b50d6 _PyEval_EvalCodeWithName + 2422
31  Python                              0x00000001061ab234 PyEval_EvalCode + 100
32  Python                              0x00000001061e88f1 PyRun_FileExFlags + 209
33  Python                              0x00000001061e816a PyRun_SimpleFileExFlags + 890
34  Python                              0x00000001062079db pymain_main + 6875
35  Python                              0x0000000106207f2a _Py_UnixMain + 58
36  libdyld.dylib                       0x00007fff5d8aaed9 start + 1
37  ???                                 0x0000000000000002 0x0 + 2

任何人都可以提供一些关于 header 这是指什么以及如何解决这个问题的建议。

Answer 1

不支持将 mp3 编码的音频作为输入格式。请使用具有 16 位样本、16 kHz 采样率和单声道 (Mono) 的 WAV(PCM) 文件。

Answer 2

默认的音频流格式是 WAV（16kHz 或 8kHz、16 位和单声道 PCM）。除了 WAV / PCM 之外，还支持下面列出的压缩输入格式。

但是如果你使用C#/Java/C++/Objective C并且你想使用.mp3等压缩音频格式，你可以处理它通过使用 GStreamer

有关详细信息，请参阅此 Microsoft 文档。

https://docs.microsoft.com/en-us/azure/cognitive-services/speech-service/how-to-use-codec-compressed-audio-input-streams

Answer 3

我想没有官方方法可以使用不同格式（mp3 或不同帧率）的 SDK 我想使用能够使用任何类型的音频文件输入的 Azure 方法

到现在为止我都是用我自己编的方法来处理这个问题，先转换适当的文件，完成我的工作后再删除它。原始文件保留：

对于python:

fname_buf = fname
fname = self.AudioFileAdjust(fname,'test-it') 

# Do somethings

if fname_buf != fname:
self.AudioFileAdjust(fname,'remove')

子函数AudioFileAdjust（我正在使用pydub和pyaudio）：

def AudioFileAdjust(self,fname,states=''):
    '''
    check audio file format and if not appropriate create new buffer audio for use
    '''
    if states == 'remove':
        os.remove(fname)
    else:
        # if the file format not useful for Azure, first need to change -> fr: 16000 must be
        audio_file = au.ReadAudioFile(fname)
        if audio_file.frame_rate != int(16000):
            #print('[Commend] changing the FrameRate')
            audio_file_e = au.SetFramerate(audio_file,int(16000))
            #change fine name for use
            fname2 = fname.split(".")[0] + "_Conv_2" + ".wav"  #without wav firstly and add additional 
            au.ExportAudioFile(audio_file_e,fname2)
            #print('new file name: ', fname)
            fname = fname2
    return fname

Microsoft Speech to Text Python SDK SPXERR_INVALID_HEADER 问题

Microsoft Speech to Text Python SDK SPXERR_INVALID_HEADER issue

python-3.x

speech-recognition

speech-to-text

microsoft-cognitive