DIVERGENCE OPTIMIZATION IN NONNEGATIVE MATRIX FACTORIZATION WITH SPECTROGRAM RESTORATION FOR MULTICHANNEL SIGNAL SEPARATION

In this paper, we address an optimization issue for the divergence in supervised nonnegative matrix factorization with spectrogram restoration, which has been proposed for addressing multichannel signal separation. This method separates non-target components and reconstructs some missing data caused by preceding spatial clustering via supervised basis extrapolation. In our previous study, we only used a limited type of divergence, whereas the divergence selection is essential. Therefore, we extend this method to a more generalized form and give a theoretical analysis of the divergence optimization, where we reveal the trade-off between separation and extrapolation abilities.