Question

Google语音识别功能只能用几秒钟音频。所以我将音频文件拆分为块。这是分裂音频的类。

class Split_audio():
    def __init__(self):
        """
        Constructor
        """
    def create_folder(self,audio):
        """
        Create folder for chunks
        """
        #name of the folder: exemple audio file's name = test.wav ==> folder's name = test
        pos=audio.get_nameAudioFile()
        pos=pos.rfind('.')
        folder=audio.get_nameAudioFile()[0:pos]
        #if folder exist overwrite
        if os.path.exists(folder):
            shutil.rmtree(folder)
        #create folder
        os.makedirs(folder)
        return folder
    def split(self,audio,silence_thresh=None, min_silence_len=500):
        """
        Split audio file on silence
        """
        sound_file = AudioSegment.from_wav(audio.get_nameAudioFile())
        if silence_thresh==None:
            silence_thresh=int(sound_file.dBFS)-19
        audio_chunks = split_on_silence(sound_file, silence_thresh=silence_thresh, min_silence_len=min_silence_len)
        return audio_chunks
    def export(self,audio,path_folder=None):
        """
        Export chunks on wav's file
        """
        audio_chunks=self.split(audio)
        if path_folder==None:
            path_folder=self.create_folder(audio)
        for i, chunk in enumerate(audio_chunks):
            out_file = "chunk{0}.wav".format(i)
            path="%s/%s" %(path_folder,out_file) 
            chunk.export(path, format="wav")

我的结论是google_recognize输出的质量取决于silence_thresh和min_silence。在对3种不同的音频进行测试之后，我将值设置为silent_thresh =音频的dbfs - 19和min_silence = 500ms。 1个月后，我重新测试了相同音频的代码。哎呀我的成绩单与第一篇完全不同。这是两个结果： First result second result。有什么建议吗？

为什么Google语音识别API会返回转换后的音频文本的一部分

0 个答案: