A novel semi-blind source separation framework towards maximum signal-to-interference ratio

Jianjun Gu, Dingding Yao, Junfeng Li, Yonghong Yan

Abstract: Highlights•A new semi-blind source separation method termed minimum variance independent component analysis (MVICA).•A deep neural network-supported implementation of MVICA.•The relationships between MVICA and the existing speech separation approaches.•MVICA outperforms the state-of-the-art BSS techniques and beamformers.