Amphion icon indicating copy to clipboard operation
Amphion copied to clipboard

Published 20 hours ago • verified publisher icon open-mmlab

[Help]: about VALLE training

Open walker-hyf opened this issue 1 year ago • 2 comments

After training valle on libritts, we noticed a drop in the Train Loss and a slight increase in the Valid Loss. Plus, when we trained AR and NAR models for 20 Epochs each, the synthesized speech quality wasn't great, and unfortunately, even after 100 Epochs of team training, the results were disappointing. Did the team encounter an increase in ValidLoss during the training process? I only made use of lexicon without making any other changes.

TensorBoard: image

INFO: image

ckpts.json image

walker-hyf avatar Jan 15 '24 09:01 walker-hyf

Could you please give some your samples?

lmxue avatar Jan 16 '24 13:01 lmxue

Hi @walker-hyf , Could you give us some of your samples to help us identify your problems?

HarryHe11 avatar Feb 16 '24 09:02 HarryHe11