Hi folks,

so here's one of latest additions to local AI music stuff generation - YuE

generating vocals based on lyrics text input + (optionally) instrumental audio track,
not sure if I got things running optimized, 1 minute result takes about 1 hour processing on 16GB RTX A4000, and quality is not that great (would require even more beefy gpu)


here are few examples (same lyrics prompt, different instrumentals made with UVR5):



you can guess which source songs were used

 

I was thinking that as a producer I must record vocals singer or myself on my home studio?

 

Why does this sound so bad, even the instrumentals? I never got such bad results using UVR.

 

Other than being low quality, it sounds impressively human. I hope it doesn’t get better, but it will of course.

 

Mono?

 

"take a look around" vocal sounds less realistic to me than the next 2, and then "trippin at the disco" sounds more AI generated like the first one. Maybe the more lo-fi or organic the backing music is, the more the vocal sounds unnatural with it. If someone guessed it was caused by Autotune, it wouldn't be that much of a surprise.