Performance comparison of several adaptive schemes for microphone array beamforming

Abstract

The use of microphone array can be a valuable resort for robust speech acquisition. This paper describes a comparative study of the performance achieved by three different schemes of adaptive beamforming with microphone arrays: linear constrained beamforming (Frost [1]), AMNOR system [2] and Affes system [3]. In general, these adaptive schemes are suitable for speech enhancement systems in noisy and reverberant environments, and in the presence of other interfering speech sources (multi-talkers environments). This analysis also includes the classical delay and sum (DS) scheme as a reference for comparative purposes. In order to achieve a homogeneous comparison of the performance, several experimental environments was simulated to evaluate the different schemes, consisting of a desired speech source, an interfering speech or white noise source and the presence of reverberation and omnidirectional noise. This analysis includes simulations of uniform linear arrays, harmonically nested arrays, as well as planar configurations (circular and L-shaped arrays). The Affes system shows be the most suitable system in presence of directional interferences, reverberation and omni-directional noise.