Music Signal Separation Based on Supervised Nonnegative Matrix Factorization with Orthogonality and Maximum-Divergence Penalties
Daichi Kitamura, Hiroshi Saruwatari, Yagi Kosuke, Kiyohiro Shikano, Yu Takahashi, Kazunobu Kondo · IEICE Transactions on Fundamentals of Electronics Communications and Computer Sciences · 2014
In this letter, we address monaural source separation based on supervised nonnegative matrix factorization (SNMF) and propose a new penalized SNMF. Conventional SNMF often degrades the separation performance owing to the basis-sharing problem. Our penalized SNMF forces nontarget bases to become different from the target bases, which increases the separated sound quality.