Speech enhancement based on sparse representation using universal dictionary

An effective approach to speech enhancement based on sparse representation is proposed. More specifically, a universal dictionary was trained on many clean speech utterances for alternative speech representation by adopting the K-SVD algorithm. While the universal dictionary could be processed beforehand, a lot of time consumption would be saved during denoising procedure. Then orthogonal matching pursuit (OMP) algorithm was employed to reconstruct the target speech over the universal dictionary. Experimental results show that the proposed approach achieves better or similar perceptual evaluation of speech quality (PESQ) scores and output SNR compared to other conventional methods in a wide range of input SNR.