hardy tiger
@white rain You were absolutely right, my first model using Applio (KLM 5.0 & your experimental RefineGAN - 500 Epochs) came out like autotune robot trash at every single Epoch save, however the voice model produced with the same dataset using this older pretrain with this older software https://github.com/RVC-Project/Retrieval-based-Voice-Conversion-WebUI came out sounding very accurate. Why did these newer pretrained models & software with higher epochs severely underperform? Could it be I was using the wrong type of pretrain model? What newer pretrains would you recommend?
For reference, I am voice cloning a very animalistic, sadistic and gutteral sounding creature from the game Halo 2, specifically Brute Bloodthirsty, and there are more niche voices like the Brute I would like to clone as well.
https://www.youtube.com/watch?v=-McTL1Bznik
I am going to throw down some links to the audio so you can hear how nice this came out.
"Brute Bloodthirsty - Built-in Pretrain - 50 epochs - RVC-Project"
https://jumpshare.com/s/VwS3GI3BXv15bjosK56v
https://jumpshare.com/s/w34w3kciRsMSH3ygI1RX
https://jumpshare.com/s/IpVQYWIu3A9e0V8CXmrU
"Brute Bloodthirsty - Exp RefineGAN 44khz + KLM 5.0 - 500 epochs - Applio"
Every epoch save sounds pretty similar to that audio above.
white rain
smoky meadow