An Approach of Speech Enhancement by Sparse Code Shrinkage

This paper presents an approach of speech enhancement by sparse code shrinkage (SCS), which mainly utilizes the sparsity of the components obtained by transmitting the noisy data through a basis functions matrix. Using the maximum likelihood estimation (MLE), the sparse components can be nonlinearly de-noised. Finally by the inverse transformation and reconstruction, the estimated original speech can be obtained. The basis functions matrix reflects statistical character of the speech data themselves, so the SCS method is of great rationality. Computer simulation result exhibits that the SCS has excellent speech enhancement effect. In comparison with wavelet shrinkage, SCS shows prominent performance