Robotic noise on audio
Likkkez opened this issue · comments
I'm trying to fine-tune the 24kHZ model but it seems that the more I train it the worse it becomes. Here's how the audio sounds now:
out.mp4
I have a custom dataset with a few speakers. Any idea how to get better results? Would adding in the original VCTK dataset to the mix make it better?
maybe the custom dataset is too small and so leads to overfitting. mix with adding original vctk data may help.
maybe the custom dataset is too small and so leads to overfitting. mix with adding original vctk data may help.
Should I restart finetuning after i add VCTK or just continue from this checkpoint?
i would prefer restart after add vctk
i would prefer restart after add vctk
I added the vctk and trained it more from an earlier checkpoint. I think it cleared up pretty nicely! Thanks for all the help!