Sparse representation and dictionary learning are popular techniques for linear inverse problems such as denoising. In this paper greedy adaptive dictionary learning algorithms are applied to the speech denoising problem, and its performance is analyzed for different noisy speech signal taken from Noises data base. In the methods of denoising, an over-complete dictionary is learned using variety of K-singular value decomposition (K-SVD) algorithm and are compared. Clean speech signal in obtaining from the noisy speech signal using the sparse signal and the dictionary. The variants of the K-SVD that are described in this paper are the simple K-SVD, generalized K-SVD, Label Consistent (LC) K-SVD and Locality preserving (LP) K-SVD and Rotate-KSVD. And it is found that though K-SVD algorithm has better SNR when compared to other algorithms the computational time taken by the R-SVD less than other algorithms with marginal difference in signal to noise ratio (SNR).
Volume 11 | Issue 8