An improved MAP-based speech enhancer for high sound quality in automobile environment

In recent years, demand for speech enhancement under a low SNR environment in the traveling automobile becomes higher. Accordingly, we have proposed a speech enhancement method that is a MAP-based speech enhancer combined with a null beamformer. Although this method gave a better noise suppression capability, it did not give a sufficient speech quality for practical use. In this paper, we introduce three novel techniques with respect to this method, i.e., gain smoothing, concise voice activity detection, and small gain adjustment. These techniques strongly reduce residual noise and musical noise, especially in the case of a low SNR. From computer simulations, we confirmd that the proposed method is effective in the automobile environment.