【Dual-Path-RNN-Pytorch源码分析】AudioReader

AudioReader 是Dual-Path-RNN-Pytorch源码中用来读入scp文件的工具类，它的输入是一个scp文件的全路径。

有关scp文件，它是一个文件与文件路径的记录文件。 内容样例如下

1.wav  /local/file/path/1.wav
2.wav  /local/file/path/2.wav
3.wav  /local/file/path/3.wav
4.wav  /local/file/path/4.wav
5.wav  /local/file/path/5.wav
6.wav  /local/file/path/6.wav

AudioReader 通过torchaudio读入文件中的每一个音频文件，这里注意一下，有些同学可能安装完torchaudio后但是会有如下错误：
No audio I/O backend is available
torchaudio需要一个具体“做事”的软件（backend）去执行载入音频文件，所以需要安装它。

torchaudio在windows下的backend默认为soundfile
使用pip install SoundFile 即可以安装
torchaudio在linux下的backend默认为sox

所以它的读操作如下，即使用了torchaudio.load

def read_wav(fname, return_rate=False):
    '''
         Read wavfile using Pytorch audio
         input:
               fname: wav file path
               return_rate: Whether to return the sampling rate
         output:
                src: output tensor of size C x L 
                     L is the number of audio frames 
                     C is the number of channels. 
                sr: sample rate
    '''
    src, sr = torchaudio.load(fname, channels_first=True)
    if return_rate:
        return src.squeeze(), sr
    else:
        return src.squeeze()

AudioReader 类本身可以理解为一个字典（dict），它的内部实现形式为

{
	'1.wav': '/local/file/path/1.wav',
	'2.wav': '/local/file/path/2.wav',
	'3.wav': '/local/file/path/3.wav',
	'4.wav': '/local/file/path/4.wav',
	'5.wav': '/local/file/path/5.wav',
	'6.wav': '/local/file/path/6.wav'
}

但是实际通过字典返回时，是把对应的文件load完返回数据的。这样方便调用者使用。
同时，它扩展了字典key值，允许使用int来索引。总结AudioReader不复杂，但是设计上还是不错的。

下面是全部源码，请参考

def read_wav(fname, return_rate=False):
    '''
         Read wavfile using Pytorch audio
         input:
               fname: wav file path
               return_rate: Whether to return the sampling rate
         output:
                src: output tensor of size C x L 
                     L is the number of audio frames 
                     C is the number of channels. 
                sr: sample rate
    '''
    src, sr = torchaudio.load(fname, channels_first=True)
    if return_rate:
        return src.squeeze(), sr
    else:
        return src.squeeze()


def write_wav(fname, src, sample_rate):
    '''
         Write wav file
         input:
               fname: wav file path
               src: frames of audio
               sample_rate: An integer which is the sample rate of the audio
         output:
               None
    '''
    torchaudio.save(fname, src, sample_rate)


class AudioReader(object):
    '''
        Class that reads Wav format files
        Input as a different scp file address
        Output a matrix of wav files in all scp files.
    '''

    def __init__(self, scp_path, sample_rate=8000):
        super(AudioReader, self).__init__()
        self.sample_rate = sample_rate
        self.index_dict = handle_scp(scp_path)
        self.keys = list(self.index_dict.keys())

    def _load(self, key):
        src, sr = read_wav(self.index_dict[key], return_rate=True)
        if self.sample_rate is not None and sr != self.sample_rate:
            raise RuntimeError('SampleRate mismatch: {:d} vs {:d}'.format(
                sr, self.sample_rate))
        return src

    def __len__(self):
        return len(self.keys)

    def __iter__(self):
        for key in self.keys:
            yield key, self._load(key)

    def __getitem__(self, index):
        if type(index) not in [int, str]:
            raise IndexError('Unsupported index type: {}'.format(type(index)))
        if type(index) == int:
            num_uttrs = len(self.keys)
            if num_uttrs < index and index < 0:
                raise KeyError('Interger index out of range, {:d} vs {:d}'.format(
                    index, num_uttrs))
            index = self.keys[index]
        if index not in self.index_dict:
            raise KeyError("Missing utterance {}!".format(index))

        return self._load(index)