search

jik876 / hifi-gan

HiFi-GAN: Generative Adversarial Networks for Efficient and High Fidelity Speech Synthesis
MIT License
1.92k stars 506 forks source link

Error with fine_tuning as True with mel generated from Glow-TTS #112

Open ankurdhuriya opened 2 years ago

ankurdhuriya commented 2 years ago

Screenshot from 2022-02-18 16-14-44 getting error - RuntimeError: stack expects each tensor to be equal size, but got [8192] at entry 0 and [8129] at entry 12

Do anyone have a solution to do it ?

rishikksh20 commented 2 years ago

some files are smaller than segment size of 8192 data points, remove the files which are smaller than 1 sec.

huiofficial commented 1 year ago

some files are smaller than segment size of 8192 data points, remove the files which are smaller than 1 sec.

I ran into the same problem with files much larger than 1 sec, and I found out the padding of mel and audio in meldataset.py is negative in my case, I think padding with negative doesn't make any sense here. Should I change it into the max segment size with my data?