humble walrus
i actually trained this in applio using crepe (its mangio-crepe, because applio's crepe has hop length).
this is a retrain of @tribal silo 's stirling core 600 epoch model, at the same time a model that i wanted to do because i like stirling.
samples were made using rmvpe, so i'd recommend using rmvpe for inference to get better results.
trained on a 4-minute and 24-seconds dataset ripped directly of the game using gcfscape (rx de-ess, mouth de-click, resample + eq, noise gate).
pitch extraction: mangio-crepe
hop length: 8
steps: 4.48k
batch size: 6
pretrain: original v2 / 32k
precision: fp16
cv. Will Neal
please don't forget to credit me when you use this model. also, don't re-upload any of my models without my permission.
burnt geyser
frosty forge
