Two-Microphone Noise Reduction Using Spatial Information-Based Spectral Amplitude Estimation

Abstract

SUMMARY Traditionaltwo-microphonenoisereductionalgorithmstodeal with highly nonstationary directional noises generally use the direc-tion of arrival or phase difference information. The performance of thesealgorithms deteriorate when diffuse noises coexist with nonstationary di-rectional noises in realistic adverse environments. In this paper, we presenta two-channel noise reduction algorithm using a spatial information-basedspeech estimator and a spatial-information-controlled soft-decision noiseestimator to improve the noise reduction performance in realistic non-stationary noisy environments. A target presence probability estimatorbased on Bayes rules using both phase difference and magnitude squaredcoherenceisproposedforsoft-decisionofnoiseestimation,sothattheycanshare complementary advantages when both directional noises and diffusenoises are present. Performances of the proposed two-microphone noisereduction algorithm are evaluated by noise reduction, log-spectral distance(LSD) and word recognition rate (WRR) of a distant-talking ASR systemin a real room’s noisy environment. Experimental results show that theproposed algorithm achieves better noises suppression without further dis-torting the desired signal components over the comparative dual-channelnoise reduction algorithms.