Oliver Sykes of Bring Me The Horizon

Description

Dataset and train by But you can credit my channel Redemption Music This model is good at screaming and distortion. The more Feature Retrieval number you use, the more is the chance of high notes to switch into scream or growl. Don't forget to mix harvest and crepe versions when glitching. Final result is not really satisfying, because you will not find clear acapellas. Dataset is 11:30 min without silences. I think low register could be better, if I added more songs with quiet vocal. If somebody is enthusiastic enough, you can train new version yourself or make quiet clear vocal version. Because I have other plans for now. 48khz trained on same dataset, so this versions are almost identical. 40khz 600 epoch: 48khz 350 epoch: