another phase connect model yipiiii woohoo
trained on roughly 16 minutes of data from her stream
(xminus mel roformer + uvr de-reverb (the new x-minus one, not the de-echo-de-reverb one) + de-click)
pitch extraction: rmvpe
steps: 14.9k/15k
batch size: 7
pretrain: klm 4.1 / 32k
huggingface: https://huggingface.co/sxndypz/rvc-v2-models/resolve/main/ayase_yuu.zip