paper current
first time making a vtuber model in a while. sorry if the voice model's a bit rough, i tried my best to clean the dataset.
trained on 12 minutes of her speaking from her vgm stream (unwa's mel-roformer v2 + uvr de-reverb + spectral de-noise + dialogue isolate + eq-ing)
pitch extraction: rmvpe
steps: 18.8k steps
batch size: 7
pretrain: klm 4 hfg / 40k
hf: https://huggingface.co/sxndypz/rvc-v2-models/resolve/main/eepysleepy.zip
bitter dagger
pallid moth