First-order differential microphone array for robust speech enhancement

Abstract

This paper presents a novel speech enhancement method based on first-order differential microphone array (DMA) system. The enhancement algorithm is implemented in three steps. In the first step, a time-delay of arrival (TDOA) estimator is used to determine the time difference between two microphone signals, and the intrinsic delay parameter of DMA is adjusted accordingly. In the second step, DMA speech enhancement algorithm, including low frequency compensation based least mean square (LFC-LMS) filtering and correlation detection based voice activity detection (CD-VAD), is utilized to eliminate background noise from desired speech. In the last step, DMA output will be filtered by a post-filter in order to remove residual noise, if necessary. Experiments with both simulated and real environmental data indicate that the time-delay can be robustly estimated by the TDOA estimator, and the proposed system performs well and robustly even in adverse noisy environment when the input SNR is lower than 0 dB.