Noisy Speech Enhancement Using Soft Thresholding on Selected Intrinsic Mode Functions

In this paper, a new speech enhancement method is introduced. It is essentially based on theEmpirical Mode Decomposition technique (EMD) and a soft thresholding approach applied onselected modes. The proposed method is a fully data driven approach. First the noisy speechsignal is decomposed adaptively into intrinsic oscillatory components called Intrinsic ModeFunctions (IMFs) by using a time decomposition called sifting process. Second, selected IMFsare soft thresholded and added to the remaining IMFs with the residue to reconstitute theenhanced speech signal. The proposed approach is evaluated using speech signals fromNOISEUS database corrupted with additive white Gaussian noise. Our algorithm is compared toother state of the art algorithms.