International Journal of Radio, Space and Aerospace Engineering

Authors: Nayana C.G, Shruthi S Prabhu, Parameshachari B.D, Rudresh M.D

Abstract: From last few decades many frequency domain speech denoising algorithms have been proposed for the speech enhancement. Purpose of speech enhancement is to improve the quality of speech which is most prominent in telecommunication applications. The main objective of this paper is to develop a time domain speech denoising algorithm which offers superior performance compared to previous approaches. This paper demonstrates the time domain methods of speech denoising called Empirical Mode Decomposition (EMD). EMD decomposes speech signal corrupted by noise into band limiting signals called intrinsic mode functions. Each of the IMF is thresholded directly; on the other hand the IMF’s are divided into frames then thresholded for noise removal. One of the drawbacks of EMD is mode mixing, which is often a consequence of signal intermittency. The mode mixing can be elevated by Ensemble Empirical Mode Decomposition (EEMD) which is the modification of EMD. The performance of the EMD and EEMD algorithm are evaluated for wide range of input SNR’s and good efficiency is derived for EEMD based framed thresholding for speech enhancement.