Gain factor linear prediction based decision-directed method for the a priori SNR estimation

The performance of a noisy speech enhancement algorithm depends mainly on the accuracy of the a priori signal-to-noise ratio (SNR) estimate. The decision-directed (DD) algorithm for estimating the a priori SNR has received lots of attention due to its good performance in eliminating the musical noise and the low computational complexity. However, this algorithm has a serious problem in that the estimation result of the a priori SNR tracks the shape of the instantaneous SNR with one frame delay, which leads to the degraded quality of enhanced speech signal. To remove this drawback, our paper proposes a gain factor linear prediction based DD approach for the a priori SNR estimation. Firstly, we analyze the statistical dependence between the successive gain factors, and develop an adaptive scheme to predicate the current gain factor using the gain factors at previous fames. Then, the predicated gain factor at the current frame is mapped with the current noisy speech to compute the first component of the DD method. The advantage of our approach is that it does no longer consist of any priori information about the estimated a priori SNR at previous frame and effectively increases the tracking sensitivity to speech onsets. In the speech enhancement simulation experiments, the proposed method is shown to bring significant improvement as compared to the conventional DD method and its modified schemes.