Music Source Separation with Generative Flow

Ge Zhu, Jordan Darefsky, Fei Jiang +2

Fully-supervised models for source separation are trained on parallel mixture-source data and are currently state-of-the-art. However, such parallel data is often difficult to obta…