Description

made by me PURE edition 43K - dataset length: 23mins - trained on clean acapellas/vocals from: interviews, open mics, live freestyles MIXED edition 64.3K - dataset length: 35 mins - trained on everything from PURE edition + songs with least amount of voice processing (556, highriser, etc.) NOTE: MIXED edition sounds more aggressive than PURE but has more artifacts and noise in the resulting audio PURE model model