Microphone array post-filter using input-output ratio of beamformer noise power spectrum

Abstract

A generalised Wiener post-filter is proposed to enhance the beamformer output in a microphone array for automatic speech recognition in mobile robot application. Conventional Wiener post-filters are known to perform poorly on suppressing low frequency noise. Proposed is a new post-filter fitting into a generalised Wiener model to reduce low frequency noise more effectively for the purpose of realising robust speech recognition. Experimental results confirm that the proposed post-filter shows significant improvement in terms of speech quality and recognition performance.