A Variable Momentum Factor Algorithm for a priori SNR Estimation in Speech Enhancement

The estimation of the a priori signal-to-noise ratio (SNR) is a very significant issue for many speech enhancement algorithms. The widely-used decision-directed (DD) algorithm largely depresses the musical noise, but the estimated a priori SNR suffer from one frame delay which results in the degradation of speech quality. In this paper, we propose a novel algorithm to a priori SNR estimation which solves the above problem while keeping the advantage of the DD approach. First, a momentum term is added and incorporated into the traditional DD approach to accelerate the tracking speed for the a posteriori SNR. Then a self-adaptive momentum factor is achieved in the minimum-mean-squared-error (MMSE) sense to improve the allover performance of the proposed algorithm. Simulation experiment results show that our proposed algorithm brings significant improvement compared to the DD and fixed momentum factor algorithms under various noisy types and levels.