-# be warned that this model has some ringing that i couldn't fix it's not the pretrain's fault ;-;
also testing out klm kpu
trained on (almost) 27 minutes of her singing from a karaoke stream
(cleaned using hyperace v2 voc, anvuew karaoke, anvuew de-reverb, some eq, resampled to 32k)
-
pitch extraction: rmvpe
steps: 21.2k
batch size: 6
pretrain: klm kpu II / 32k
