Fixed Beamformer Design Using Polynomial Eigenvalue Decomposition

Array processing is widely used in many speech applications involving multiple microphones. These applications include automatic speech recognition, robot audition, telecommunications, and hearing aids. A spatio-temporal filter for the array allows signals from different microphones to be combined desirably to improve the application performance. This paper will analyze and visually interpret the eigenvector beamformers designed by the polynomial eigenvalue decomposition (PEVD) algorithm, which are suited for arbitrary arrays. The proposed fixed PEVD beamformers are lightweight, with an average filter length of 114 and perform comparably to classical data-dependent minimum variance distortionless response (MVDR) and linearly constrained minimum variance (LCMV) beamformers for the separation of sources closely spaced by 5 degrees.

1. Introduction

Array processing is widely used in many speech applications involving multiple microphones. These applications include automatic speech recognition [1], robot audition [2], telecommunications and hearing aids [3]. Beamforming, or the design of a spatio-temporal filter for the array, combines signals from different microphones to extract the desired signal arriving from a specific direction [4]. Processing these extracted signals instead of the microphone signals usually improves the application performance.