Microphone array processing using reflections constructively

Abstract

A method for speech enhancement in a reverberating environment using microphone arrays is presented. It is defined in the time-domain, based on an acoustic signal model formulated using image sources. The method uses the output of a delay and sum (DAS) beamformer steered towards a source together with the outputs of multiple DAS beamformers steered towards the reflections. The outputs are delayed and added together so that the direct signal and the reflections are added coherently. A minimum power/distortionless response (MVDR)-formulation for weighting of the reflections is also presented. The method requires that the directions of arrival (DOA) and time differences of arrival (TDOA) for the reflections are known. In addition the MVDR formulation requires that the relative reflection strengths are known approximately. Simulations and measurements in a real room show that the method suppresses reverberations, interfering sources, and microphone self noise. The direct signal is not distorted. The method has improved performance compared to the conventional DAS beamformer given that the reflections used in the beamformer are strong.