Deep-Learning Vocal Remover

Introduction

My own easy-to-understand implementation of the paper Jansson et al., "Singing Voice Separation with Deep U-Net Convolutional Networks" using PyTorch and librosa.

Usage

Training

Put audio files with instrument-only track on the left channel and mixed (with vocal) track on the right channel into the data directory.
Run train.py

Inference

Specify input media in inference.py.
Run inference.py
The result will be saved as result.wav.