Noise estimation based on an adaptive smoothing factor for improving speech quality in a dual-microphone noise suppression system

In this paper, we present a noise estimation algorithm using an adaptive smoothing factor based on the Teager energy operator (TEO) energy ratio instead of a voice activity detector (VAD) for a dual-microphone noise suppression system. Conventionally, a dual-microphone noise suppression system consists of two independent systems: a beamformer or adaptive noise canceller (ANC) as a front-end processor and a single-channel-based noise suppression system as a postprocessor. In this paper, we propose to treat the two independent systems as two components of a single system. This can be accomplished by using the knowledge learned from the dual-microphone front-end processor to control the noise power spectrum estimation of the single-channel-based postprocessor. Under severe background noise conditions, this method significantly reduces the speech distortion caused by the single-channel postprocessor. The proposed algorithm is of relatively low complexity, minimizes VAD detection error, and features robustness to both high and low input signal SNRs.