A Multi-Channel Noise Estimator Based on Improved Minima Controlled Recursive Averaging for Speech Enhancement

Abstract

This article introduces an extension of the improved minima-controlled recursive averaging noise estimation from single to multi-channel speech enhancement systems. With the spatial information of microphone array signals being fully exploited, more accurate estimate of the noise spectrum can be obtained over the single-channel counterpart. Computer simulation demonstrates superior performance of the proposed noise estimator in terms of noise tracking performance and noise estimation error. Furthermore, the use of the proposed technique with the multi-channel Wiener filter yields improved signal-to-noise ratio and speech distortion.