Tongue-tied voices

[gazsiazasz]

I tried to train a model using English voices from a game, then I tried to infer a Hungarian voice with the model. The result was kinda low quality sounding. And the R sounds were bitten off.

Then I downloaded a a pretrained anime model from huggingfaces with 20000+ epoch. The sound quality was much better this time on the Hungarian voice file, but the R sounds were still bitten off.

Is there any way to overcome this issue?