2023.01.10 update: code below can deteriorate model performance

[Dannynis]

Hey, by commenting the code in data_utils.py you introduced a potential bug if the wav files are not equality sized

[skol101]

Could you provide link to what code exactly ?

[Dannynis]

when commenting this code: ''' lmin = min(c.size(-1), spec.size(-1)) spec, c = spec[:, :lmin], c[:, :lmin] audio_norm = audio_norm[:, :lminself.hop_length] _spec, _c, _audio_norm = spec, c, audio_norm while spec.size(-1) < self.spec_len: spec = torch.cat((spec, _spec), -1) c = torch.cat((c, _c), -1) audio_norm = torch.cat((audio_norm, _audio_norm), -1) start = random.randint(0, spec.size(-1) - self.spec_len) end = start + self.spec_len spec = spec[:, start:end] c = c[:, start:end] audio_norm = audio_norm[:, startself.hop_length:end*self.hop_length] '''

[skol101]

So, you're saying this code must be UNcommented?