I've made this model ***THREE TIMES*** and yet somehow I wasn't really proud with any of them, so I made one final version of this model using my newest method of training. https://huggingface.co/Xhepyxopila/MarioRVCModels/resolve/main/BirdoGBAv2.zip

i'Ll ReMeMbEr ThIs

I get the audio in the highest quality I possibly can (in this case it's ripped straight from the game), I splice all the clips together in FL Studio, make sure the volume is consistent between clips, then export it as a 32-bit WAV and upload that in a .zip to Google Drive

Then in the Collab I set it to be a 48k model, set the pitch extraction thing to Mangio-Crepe with the number set to 10

Then I just set the Epoch count to something high like 500-1000 and let it do it's thing while checking for overtraining

After retraining some of my older models doing that i've found it usually results in more accurate sounding audio