Speech enhancement in EMD domain using spectral subtraction and Wiener filter Kais KHALDI *#1 , Haifa TOUATI $2 * Unité Signaux et Sytèmes, ENIT, ElManar University BP 37, Le Belevedère, 1002 Tunis, Tunisia # College of Science and Arts-Tabarjal, Al Jouf University P.O.Box 2014, Al-Jouf , Skaka,42421, KSA $ ISSAT, University of Gafsa, BP 116, Sidi Ahmed Zarrouk , 2112 Gafsa, Tunisia 1 kais.khaldi@gmail.com 2 haifatwati1991@gmail.com Abstract— This paper proposes a technique in Empirical Mode Decomposition (EMD) domain to enhance the signal. The noisy signal is decomposed, by EMD, into approximation and detail which are filtered separately using spectral subtraction and Wiener filter. Therefore, the main idea of the proposed approach is to filter the shorter scale IMF (detail) by Wiener filter, which are noise dominated, and filter the approximation using spectral subtraction technique. In fact, the filtering of the approximation by the same filter (Wiener) will introduce signal distortion rather than a noise reduction. Thus, the performance of this method is to construct linearly the original signal without loss of the useful information. The study is limited to signals corrupted by additive white Gaussian noise. Keywords— Empirical Mode Decomposition, Wiener filter, Spectral Subtraction , Speech enhancement, detail, approximation. I. INTRODUCTION Speech signal noise reduction is a well known problem in signal processing. Particularly, in the case of additive white Gaussian noise a number of filtering methods has been proposed[1]-[2]. However, these methods are not effective when the noise estimation is not possible. To overcome these difficulties, nonlinear methods have been proposed and especially those based on Wavelets thresholding [2]-[3]. The idea of wavelet thresholding relies on the assumption that signal magnitudes dominate the magnitudes of the noise in a wavelet representation, so that wavelet coefficients can be set to zero if their magnitudes are less than a pre-determined threshold [3]. A limit of the wavelet approach is that the basis functions are fixed, and thus do not necessarily match all real signals. Recently, a new temporal signal decomposition method, called Empirical Mode Decomposition (EMD), has been introduced by Huang et al. [4] for analysing data from non stationary and nonlinear processes. The major advantage of the EMD is that the basis functions are derived from the signal itself. Hence, the analysis is adaptive in contrast to the traditional methods where the basis functions are fixed. In our previous works [5]-[6], the denoising method is based on the filtering of all IMFs extracted from the noisy signal by the same filter. However, the longer scale IMFs (low- and medium-frequency components) corresponding to the most important structures of the signal is signal dominated. Therefore, filtering of these IMFs will introduce signal distortion rather than a noise reduction [7]. The basic idea of the proposed method is to filter the shorter scale IMF (detail) by Wiener filter, which are noise dominated, and filter the approximation using spectral subtraction technique. In fact the filtering of all IMFs by the Wiener filter generates a distortion of the signal, i.e. the filtering eliminates even the useful information. While the filtering of all IMFs by the spectral subtraction filter does not make it possible to effectively eliminate all the noise. The paper is organized as follows. Section II explains the basics of the EMD and Section III exposed the proposed method. Results are presented in Section IV, and conclusions are drawn in Section V. II. EMD BASICS The EMD decomposes a signal f(t) into a series of IMFs through an iterative process called sifting; each one, with distinct time scale [8]. The decomposition is based on the local time scale of f(t) and yields adaptive basis functions. The EMD can be seen as a type of wavelet decomposition whose subbands are built up as needed to separate the different components of f(t). Each IMF replaces the signals detail, at a certain scale or frequency band [9]. The EMD picks out the highest-frequency oscillation that remains in f(t). By definition, an IMF satisfies two conditions: • the number of extrema and the number of zero crossings may differ by no more than one; • the average value of the envelope defined by the local maxima and the envelope defined by the local minima is zero. 5th International Conference on Control Engineering&Information Technology (CEIT-2017) Proceeding of Engineering and Technology –PET Vol.32 pp.27-32 Copyright IPCO-2017 ISSN 2356-5608