Abstract

Music source separation is an important task for many applications in music information retrieval field. However, due to the complexity of the music signal t is still considered a challenging task. In this paper, we propose a novel approach extending Wasserstein generative adversarial networks to source separation task. We used the mixture signal as a condition to generate sources and applied the U-net style network for the stable training of the model. Experiments with the DSD100 dataset show the promising results with the potential of using the GANs for music source separation.