Description

EDIT: Working on a harvest retrain, since this is a lower quality dataset. Should help quite a bit. For the meantime you can use this or experiment with v1 kanye. The same Kanye RVC model, except I improved the dataset and retrained for v2. Steps I took (after getting the dataset from who made the original) - Consistent frequencies (used Phaser in RX 9 for this) - Equalized & normalized the quiet sections to match the rest of the dataset - Reverb removal on a minute or two of data - Noise gated to remove most background noise in the dataset - Most important of all, retrained using RVC v2 with crepe As a result it's much better, but I'm sure I'll be updating this at some point