Dual-microphone speech enhancement system attenuating both coherent and diffuse background noise

Abstract

In this paper, we present an adaptive dual-microphone array to suppress coherent as well as diffuse noise in disturbed speech signals. This system consists of a dual microphone array and an algorithm for processing their signals. The algorithm for target speech enhancement consists of two main algorithmic steps: segmentation and enhancement. The segmentation is realized using GCC-PHAT signal processing algorithm. This algorithm allows to detect the target speaker’s speech and extract it in a noisy environment. Further the target speaker’s speech is enhanced using an adaptive Minimum Variance Distortionless Response (MVDR) filter implemented in the frequency domain. This paper proposes a practical improvement of MVDR filter which allows better save speech of the target speaker. Experiments with real recordings demonstrate the reduction of both coherent and diffuse background noise. The main advantage of the proposed technique is simplicity of its use in a wide range of practical situations.