Retrieval-based-Voice-Conversion-WebUI icon indicating copy to clipboard operation
Retrieval-based-Voice-Conversion-WebUI copied to clipboard

In Colab, the epochs progress strangely fast, and an unfinished voice synthesis is generated.

Open mikisayakamitakihara opened this issue 2 years ago • 4 comments

In Colab, when I run training with the current ipynb notebook, the epochs progress very quickly (about 1 epoch in 6 seconds with Tesla T4 and batch size 14), and even after training for about 300 epochs, a low-quality, grainy voice synthesis weight.pth file is generated. I have over 1000 training files totaling more than 50 minutes. Is everyone else experiencing the same issue?

mikisayakamitakihara avatar Apr 15 '23 04:04 mikisayakamitakihara

There might have been too many files with low volume. I'm trying to resolve it now.

mikisayakamitakihara avatar Apr 15 '23 07:04 mikisayakamitakihara

Reduce the epoch round may fix this (ex. limit to <= 200 epochs)

fumiama avatar Apr 15 '23 07:04 fumiama

I'll give that a try as well. Thank you for the advice. After normalizing the volume, the progress of the epochs became slow, and the results improved somewhat.    I might write the results of various settings changes I make after this here.

mikisayakamitakihara avatar Apr 15 '23 08:04 mikisayakamitakihara

Ok. Looking forward to your result.

fumiama avatar Apr 15 '23 10:04 fumiama