A New Sparse Representation Algorithm for Speech Denoising

This paper proposes a new speech denoising method that uses K-SVD sparse representation algorithm. This approach is based on sparse and redundant representation over dictionary. Here, spectrogram patches are used as training samples for the initial redundant dictionary. However, since the K-SVD algorithm is limited in handling small size spectrogram, the training samples need to extend their deployment to arbitrary spectrogram sizes by defining a global spectrogram prior that forces sparsity over patches in every location in the spectrogram. Simulation experiments show that the performance of the proposed K-SVD denoising algorithm is stable, and the white noise can be effectively separated. In addition, K-SVD algorithm is a simple and effective algorithm which surpasses the redundant DCT method and Gabor dictionary. In a word, K-SVD algorithm leads to an alternative speech denoising method.