Speech Signal Compressed Sensing Based on K- Svd Adaptive Dictionary

This paper proposes a novel and successful method for speech signal compressed sensing based on KSingular Value Decomposition (K-SVD) algorithm. K-SVD is an iterative method that alternates between sparse representation of the train samples based on the current dictionary and a process of updating the dictionary atoms to better fit the speech data. The presented K-SVD algorithm is applied here for training an adaptive overcomplete dictionary which can best suit a set of given speech signals. The sparse coefficients can be obtained by conducting the Orthogonal Matching Pursuit(OMP)sparse decomposition algorithm. At last, the original signal can be reconstructed by exploiting reconstruction algorithm. The experimental results show that, compared with the traditional basis function dictionary, the proposed method has a stronger adaptability which can be availably used for speech signal sparse representiation. Moreover, the compressed sensing based on K-SVD algorithm achieves higher reconstruction accuracy and better denoising efficiency.