LCMV-based Scan-and-Sum Beamforming for Region Source Extraction

This paper addresses region source extraction (RSE), which extracts the entire sound source originating from a specified target region. To solve this problem, an LCMV-based Scan-and-Sum Beamformer (LCMV-SS-BF) that can separate a wide range of continuous regions is proposed. This method is designed by incorporating the linearly constrained minimum variance Beamformer (LCMV-BF) as a filter for continuous regions into the Scan-and-Sum Beamformer (SS-BF) as a filter for wide-area sound source separation. In addition, an incremental estimation method using the Woodbury’s formula for LCMV-BF filters is proposed to improve the processing speed of the filter computation of the LCMV-BF. Numerical simulation using convolved sound datasets showed 1) the effectiveness of the LCMV-SS-BF for surface source separation and 2) the computational cost advantage of the incremental estimation method.