<@901508003272982528>

delete your samples and provide vocal only ones

yeah?

okay

can’t risk copyrighted content

wait a minute need to do new samples

IS this why I can't find a slick rick model? About to make one if it doesn't exist. I seem to have had some pretty good results on my first real model I trained here. I used original CD audio and ripped vox from various Jedi Mind Tricks, solo and featured Vinnie Paz versus. Thanks for doing Del and Kool Keith btw!

My model turned out great for real, but I don't even know what an epoch or the other 'knobs do. I trained here and just cleaning up, normalizing and being sure there is no other stuff going on seems to have worked as well as I could hope for. Is there something to gain by doing it natively and have you found diminishing returns or regressions with bigger datasets or have any other advice? I am grabbing a bunch of intervies and his solo padcast type of videos but I imagine that they will help more for tts, I will find out, also test replacing speech. Only have tried it with rap songs..