search

insunhwang89 / StyleVC

MIT License
30 stars 3 forks source link

Is 100k enough to train the model? #7

Closed skol101 closed 2 years ago

skol101 commented 2 years ago

I find that even at 145k steps some even seen voices are subjectively quite far from the source voices.

Also training vocoder from the scratch, now at 140k step.s

insunhwang89 commented 2 years ago

It would be best if you found the appropriate parameters to fit your dataset. In my case, I conducted the experiment by fixing 100K.