Realizing speech enhancement by combining EEMD and K-SVD dictionary training algorithm

This paper presents a speech enhancement algorithm that combines the ensemble empirical mode decomposition (EEMD) and the K-singular value decomposition (K-SVD) dictionary-training algorithm together to obtain clean speech from noisy speech. The EEMD algorithm is firstly employed to obtain intrinsic mode function (IMF) components from noisy speech. The cross-correlations and autocorrelations of each IMF are calculated from the IMF components to filter out the noisy IMF components. Meanwhile, the transition IMF components are again decomposed with EEMD to further remove the noisy component. The remained original IMFs alone with the remained transition IMFs are then superimposed to generate the new noisy speech. The new noisy speech is then sparse de-composed by the K-SVD dictionary-training algorithm with an over-complete dictionary trained from clean speech. Enhanced speech is obtained by recovering the speech signal from sparse coefficient vectors. Different from the traditional speech enhancement algorithms, the algorithm enhances the noisy speech by the sparse representation of noisy speech that has been pre-de-noised with EEMD algorithm previously. Experimental results show that the algorithm achieves significant de-noising results than the traditional spectral subtraction, wavelet threshold de-noising algorithm and K-SVD dictionary-training algorithm under both low SNR situation and high SNR situation.