Instantaneous Fundamental Frequency Estimation from Speech using Fourier Decomposition Method

Abstract

Speech analysis and various speech processing applications use instantaneous fundamental frequency $(F_{0})$ of voiced speech signal as a prime acoustic parameter. The low frequency component of the voiced speech possesses most of the energy in F0 neighbourhood and its few harmonics. In this study, a novel approach is proposed to extract the instantaneous F0 component from a voiced speech signal using Fourier decomposition method (FDM), which decomposes the signal into its amplitude-frequency modulated (AM-FM) components. We also demonstrate that these derived AM-FM components, obtained due to desired frequency band decomposition property of FDM, provides the most suitable representation for voiced speech as compared to other AM-FM models available in the literature. Numerical results are presented to validate the adequacy of proposed method in estimating F0, when compared with existing algorithms based on empirical mode decomposition (EMD) and other speech-related algorithms.