def __call__(self, examples): batch_size = len(examples) mels = [example[0] for example in examples] wavs = [example[1] for example in examples] mels = batch_spec(mels, pad_value=self.padding_value) wavs = batch_wav(wavs, pad_value=self.padding_value) audio_starts = np.zeros((batch_size, ), dtype=np.int64) return mels, wavs, audio_starts
def __call__(self, samples): # transform them first if self.valid: samples = [(audio, mel_spectrogram, 0) for audio, mel_spectrogram in samples] else: samples = [self.random_crop(sample) for sample in samples] # batch them audios = [sample[0] for sample in samples] audio_starts = [sample[2] for sample in samples] mels = [sample[1] for sample in samples] mels = batch_spec(mels) if self.valid: audios = batch_wav(audios, dtype=np.float32) else: audios = np.array(audios, dtype=np.float32) audio_starts = np.array(audio_starts, dtype=np.int64) return audios, mels, audio_starts
def __call__(self, examples): mels = [example[0] for example in examples] wavs = [example[1] for example in examples] mels = batch_spec(mels, pad_value=self.padding_value) wavs = batch_wav(wavs, pad_value=self.padding_value) return mels, wavs