Empirical Mode Decomposition for adaptive AM-FM analysis of Speech: A Review

Abstract

This work reviews the advancements in the non-conventional analysis of speech signals, particularly from an AM-FM analysis point of view. The benefits of such an analysis, as opposed to the traditional shorttime analysis of speech, is illustrated in this work. The inherent non-linearity of the speech productionsystem is discussed. The limitations of Fourier analysis, Linear Prediction (LP) analysis, and the Mel Filterbank Cepstral Coefficients (MFCCs), are presented, thus providing the motivation for the AM-FM representation of speech. The principle and methodology of traditional AM-FM analysis is discussed, as amethod of capturing the non-linear dynamics of the speech signal. The technique of Empirical Mode Decomposition (EMD) is then introduced as a means of performing adaptive AM-FM analysis of speech, alleviating the limitations of the fixed analysis provided by the traditional AM-FM methodology. The merits and demerits of EMD with respect to traditional AM-FM analysis is discussed. The developments of EMD to counter its demerits are presented. Selected applications of EMD in speech processing are briefly reviewed. The paper concludes by pointing out some aspects of speech processing where EMD might be explored.