Generative Adversarial Source Separation

Cem Subakan, Paris Smaragdis

Generative source separation methods such as non-negative matrix factorization (NMF) or auto-encoders, rely on the assumption of an output probability density. Generative Adversari…