Instantaneous fundamental frequency estimation of speech signals using DESA in low-frequency region

Abstract

Instantaneous fundamental frequency (IFF) is one of the basic signal parameters of speech signals. Accurate estimation of the IFF of speech signals can be used for speaker recognition system, gender identification and emotion recognition. Various methods for estimation of the IFF of speech signals have been proposed in the literature. In this paper, we propose a new method to estimate IFF of speech signals using the discrete energy separation algorithm (DESA) in the low-frequency region. The Fourier-Bessel (FB) series expansion has been used to filter out the low-frequency region from the speech signal. The reconstructed band-limited signal from the FB coefficients is modeled by using the amplitude and frequency modulated (AM-FM) signal model. Amplitude envelope of the reconstructed band-limited signal is estimated using the DESA. The estimation of amplitude envelope function has been used for detection of glottal closure instants (GCIs). Inverse of time elapsed between two consecutive GCIs provides IFF of speech signals. Simulation results of proposed method have been compared with the IFF computed from reference differenced electroglottograph (EGG) signal. The CMU-Arctic database has been used in this work for validation of the proposed method for IFF determination. It has been observed that the proposed method provides accurate estimation of IFF of speech signals at most of time instants. © 2013 IEEE.