tsurumeso/vocal-remover

Implements a U-Net convolutional architecture trained on paired instrumental/vocal audio for source separation, with support for CPU and GPU inference. Includes Test-Time Augmentation (TTA) for improved separation quality and optional post-processing that masks instrumental output based on vocal energy detection. Built on PyTorch with a trainable pipeline allowing users to fine-tune models on custom datasets.

1,744 stars. No commits in the last 6 months.