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, examples):
ids = [example[0] for example in examples]
mels = [example[1] for example in examples]
stop_probs = [example[2] for example in examples]
ids = batch_text_id(ids, pad_id=self.padding_idx)
mels = batch_spec(mels, pad_value=self.padding_value)
stop_probs = batch_text_id(stop_probs, pad_id=self.padding_idx)
return ids, np.transpose(mels, [0, 2, 1]), stop_probs
def __call__(self, examples):
mels = []
wavs = []
starts = []
for example in examples:
mel, wav_clip, start = self.clip(example)
mels.append(mel)
wavs.append(wav_clip)
starts.append(start)
mels = batch_spec(mels)
wavs = np.stack(wavs)
starts = np.array(starts, dtype=np.int64)
return mels, wavs, starts
def __call__(self, examples):
texts = []
mels = []
text_lens = []
mel_lens = []
stop_tokens = []
for data in examples:
text, mel = data
text = np.array(text, dtype=np.int64)
text_lens.append(len(text))
mels.append(mel)
texts.append(text)
mel_lens.append(mel.shape[1])
stop_token = np.zeros([mel.shape[1] - 1], dtype=np.float32)
stop_tokens.append(np.append(stop_token, 1.0))
# Sort by text_len in descending order
texts = [
i for i, _ in sorted(
zip(texts, text_lens), key=lambda x: x[1], reverse=True)
]
mels = [
i for i, _ in sorted(
zip(mels, text_lens), key=lambda x: x[1], reverse=True)
]
mel_lens = [
i for i, _ in sorted(
zip(mel_lens, text_lens), key=lambda x: x[1], reverse=True)
]
stop_tokens = [
i for i, _ in sorted(
zip(stop_tokens, text_lens), key=lambda x: x[1], reverse=True)
]
text_lens = sorted(text_lens, reverse=True)
# Pad sequence with largest len of the batch
texts = batch_text_id(texts, pad_id=self.padding_idx)
mels = np.transpose(batch_spec(mels, pad_value=self.padding_value),
axes=(0, 2, 1))
stop_tokens = batch_text_id(stop_tokens,
pad_id=self.padding_stop_token,
dtype=mels[0].dtype)
return (texts, mels, text_lens, mel_lens, stop_tokens)
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