Speech Denoising Based on Empirical Mode Decomposition and Improved Thresholding

Abstract

The goal of this paper is to introduce a new speech enhancement algorithm using empirical-mode-decomposition (EMD). A mode selection approach and an improved thresholding technique are proposed. In the first stage, the noisy speech is adaptively decomposed into intrinsic oscillatory components called Intrinsic Mode Functions (IMFs) by means of a sifting process. In the second stage, an energy criterion, based on the minimization of the consecutive mean square error (CMSE), is applied to find the IMF from which the energy distribution of the original signal is greater than the noise. In the third stage, the previously selected IMFs are thresholded through an improved thresholding process by means of a detection function. Finally, the enhanced speech signal is reconstructed from the processed IMFs, the remaining ones and the residue. The proposed approach is evaluated using speech signals taken from the NOISEUS database corrupted with additive white Gaussian noise. Our algorithm is compared to other state of the art speech enhancement algorithms and the results are provided.KeywordsSpeech SignalEmpirical Mode DecompositionIntrinsic Mode FunctionSpeech EnhancementNoisy SpeechThese keywords were added by machine and not by the authors. This process is experimental and the keywords may be updated as the learning algorithm improves.