SyntaSpeech icon indicating copy to clipboard operation
SyntaSpeech copied to clipboard

Published 20 hours ago • verified publisher icon yerfor

Fine-Tuning approach

Open LorenzoBrugioni opened this issue 2 years ago • 0 comments

Hi, I would like to use the pretrained model on LibriTTS to adapt the model on two target speakers for which I have about 40 minutes of training data each. Could you please share how would be the approach for fine tuning it? Any modules to freeze, decreasing learning rate, if it is actually possible in your opinion with that amount of data etc.. Any info would be useful. Thanks for your work and have a good day.

LorenzoBrugioni avatar Mar 06 '23 13:03 LorenzoBrugioni