A Dual-Microphone Speech Enhancement Algorithm for Close-Talk System

While human listening is robust in complex auditory scenes, current speech enhancement algorithms do not perform well in noisy environments, even close-talk system is used. This paper addresses the robustness in dual microphone embedded close talk system by employing a computational auditory scene analysis (CASA) framework. The energy difference between the two microphones is used as the primary separation cue to estimate the ideal binary mask (IBM). We also use voice activity detection to find the noise periods, and update the separation critical value. Generalization interference locations and reverberant conditions are used to examine performance of the proposed system. Evaluation and comparison show that the proposed system outperforms other two systems on the test conditions.