vocal-remover

Why we included this project

Separating a song into vocals and instruments usually means fiddly filters or paying for an online service. vocal-remover handles it locally with a PyTorch command-line tool built around a U-Net: point it at an audio file and it writes two WAVs, one for the vocal track and one for the instrumental. It runs on CPU or GPU, and a test-time augmentation flag can improve separation quality. The repo also includes training scripts, so you can train your own model on a custom dataset of paired mixtures and instrumentals instead of relying on the bundled weights. Because it stays a focused script rather than a full studio suite, it drops neatly into batch jobs or a larger audio pipeline. Podcast editors and video producers who just need a clean stem will find it a practical, no-frills tool.