Design and Assessment of a Scan-and-sum Beamformer for Surface Sound Source Separation

Abstract

This paper proposes a novel robot audition technique for listening to a surface sound source. A beamforming technique called scan-and-sum beamformer is proposed. It performs sound source separation of general wide band surface sources distributed in the azimuth angle dimension. Because a sound source is mainly modeled as a point source in current signal processing technology, it is difficult for a robot or system with a conventional beamformer to address a surface source, though most sound sources in the real world are surface sources with a certain shape and size. The basic idea of scan-and-sum beamforming is to decompose a surface source into numerous point sources intensively distributed within a region, and then integrate results from point source sub-beamformers. In the scan-and-sum approach, a point source sub-beamformer scans with appropriate scanning density a region where target surface sources exist, and sub-beamformers are summed to separate the surface sound sources. Due to the high flexibility of the proposed beamformer, a mean square error (MSE) criterion for error analysis and cost function J to balance performance and cost are introduced, with guidelines on how to implement a scan-and-sum beamformer with tuned parameters and optimized pattern. Compared to conventional point source beamformers, beam patterns with a broader mainlobe and lower sidelobes can be achieved at acceptable cost. Numerical simulations show the proposed method improves signal-to-interference ratio (SIR) for a mixture of three surface sound sources.