StarGAN-Voice-Conversion-2 icon indicating copy to clipboard operation
StarGAN-Voice-Conversion-2 copied to clipboard

Published 20 hours ago • verified publisher icon SamuelBroughton

Training result is not good

Open lihaoyangML opened this issue 2 years ago • 0 comments

Hi, I followed the README exactly to train a StarGAN v2 model using speakers p229 p232 p236 p243 from VCTK for 200k steps, but I am getting poor results in terms of speaker similarity and audio quality. I have attached some samples generated by my trained model for your reference: https://drive.google.com/drive/folders/1u0JvrfaxkV2BRdGmc6xyEw1ue7igqKb_?usp=sharing

I would like to ask you for some guidance, if possible, on how I can improve on the training to get better performance. Below is my current training curve for your reference: image

lihaoyangML avatar Jul 07 '22 10:07 lihaoyangML