fatchord / WaveRNN

WaveRNN Vocoder + TTS

Home Page:https://fatchord.github.io/model_outputs/

Geek Repo:Geek Repo

Github PK Tool:Github PK Tool

doubts about max_offsets ?

wenyong-h opened this issue · comments

I think the max_offsets in

WaveRNN/utils/dataset.py

Lines 66 to 68 in 8fdbef1

mel_win = hp.voc_seq_len // hp.hop_length + 2 * hp.voc_pad
max_offsets = [x[0].shape[-1] -2 - (mel_win + 2 * hp.voc_pad) for x in batch]
mel_offsets = [np.random.randint(0, offset) for offset in max_offsets]

should be
max_offsets = [x[0].shape[-1] - mel_win for x in batch]
Because 2 * hp.voc_pad is already added in mel_win, I also don't understand why you minus two.

And line 68 should be changed to
mel_offsets = [np.random.randint(0, offset + 1) for offset in max_offsets]