#Which pretrain type should I use for training in this collab?
1 messages · Page 1 of 1 (latest)
depends on what you wanna do
I'd just recommend using original pretrains unless you know what you're doing
what's the difference between them
ov2 is great with very clean (DENOISED, manually) small datasets (under 10m)
rin_e3 is a "replacement" for the original ones, also requires somewhat clean datasets but gets good results
do vocals separated through phase cancellation and ran through uvr-denoise count as clean?
spectrogram of one portion
ehhhhhhh
i wouldn't bet on it
uvr denoise doesnt make things exactly clean so
ok, I suppose I'll do standard then try rin the second time
do spectral denoise
what settings
also choose ov2 if it's short