[_sys49152]

what would make these udio/suno type services better? tighter training data? exclude the muzak? how would one even go about doing that?