A computationally efficient subband noise estimation algorithm for speech enhancement

In this paper, a computationally efficient subband noise estimation algorithm for speech enhancement is presented. The noisy speech is decomposed into critical band signals and the subband noise estimation is updated by an improved minima controlled recursive averaging (IMCRA) approach. Compared to the classical IMCRA approach, the proposed subband noise estimation can exploit the human ear perceptual properties and lead to significant computational savings. Experimental results show that the proposed approach has comparable performance with lower computational requirements.