*CREPE+HYBRID TRAINING* A very experimental fork of the Retrieval-based-Voice-Conversion-WebUI repo that incorporates a variety of other f0 methods, along with a hybrid f0 nanmedian method.
MIT License
931
stars
207
forks
source link
Training too fast after first voice model created #208
Hello. I created one voice model using local Mangio, and it worked really well. Now, I wanted to create another one, but this time, the training/epochs load TOO fast, like 10 seconds every epoch and the result is really bad, I remembered that my first voice model needed like +1min for each epoch that time.
I had that SAME problem trying with WEB UI, the first model was really fine, but when I tried to create a second, the training/epochs load TOO fast the the result is really bad.
Hello. I created one voice model using local Mangio, and it worked really well. Now, I wanted to create another one, but this time, the training/epochs load TOO fast, like 10 seconds every epoch and the result is really bad, I remembered that my first voice model needed like +1min for each epoch that time.
I had that SAME problem trying with WEB UI, the first model was really fine, but when I tried to create a second, the training/epochs load TOO fast the the result is really bad.