Noise at the end of the speech

keonlee9420 / PortaSpeech

PyTorch Implementation of PortaSpeech: Portable and High-Quality Generative Text-to-Speech

MIT License

329 stars 36 forks source link

Noise at the end of the speech #13

Open yygg678 opened 2 years ago

yygg678 commented 2 years ago

hi, In the ljspeech dataset speech demo you gave, there is noise at the end of the speech. I have the same problem in the Chinese and English datasets during training. The quality of speech generation is not very good. Any suggestions

111

keonlee9420 commented 2 years ago

Hi @yyggithub , thanks for sharing your insight. I've also noted about it yet no room for fixing that. In my thought, that might be from VariationalGenerator where the Conv and ConvTrans pair applied as a frame-wise bottleneck so that the output could have such noise at the end. I surely believe that simple arithmetic can solve it, but I can't afford that for now. So it would be nice if you have time and can handle that by your side and then report them for the community. Thanks!