Speech enhancement using a joint MAP estimation of LP parameters

Codebook-based speech enhancement approach is an effective method for reducing non-stationary noise. In view of the inaccurate problem of estimating the short-term predictor parameters of the speech and noise, this paper proposes a codebook-based maximum posteriori probability (MAP) speech enhancement approach by combining MAP estimation and codebook-based method. Based on the prior information and inter-frame correlation of the short-term predictor parameters, the paper develops both memoryless and memory-based MAP predictor parameters estimators which optimally get the spectral shapes and the corresponding excitation variances. In order to further improve the accuracy of the parameters, a novel approach of estimating the excitation variances is proposed for the memory-based case. Experimental results show that, in comparison with the reference method, the proposed method can get better performance under various noise conditions.