Semi-blind source separation with multichannel variational autoencoder
share
This paper proposes a multichannel source separation method called the multichannel variational autoencoder (MVAE), which uses VAE to model and estimate the power spectrograms of the sources in a mixture. The MVAE is noteworthy in that (1) it takes full advantage of the strong representation power of deep neural networks for source power spectrogram modeling, (2) the convergence of the source separation algorithm is guaranteed, and (3) the criteria for the VAE training and source separation are consistent. Through experimental evaluations, the MVAE showed higher separation performance than a baseline method.
READ FULL TEXT