Phoneme Classification Using Modulating Features

In this paper, the robust modulating features are used from a modulated sinusoidal signal model for speech phoneme classification. The modulated sinusoidal signal model is useful for the study of non-stationary speech signals at phoneme length. The formants of the speech phoneme are extracted into the individual formants using the Fourier-Bessel analysis. The parameter estimation of the individual formants is done by analyzing the amplitude envelope and the instantaneous frequency of the spoken phoneme, which was extracted using the energy separation algorithm. For the parametric representation of the non-stationary signal, the most important features of the modulated signal models are amplitude, carrier frequency, modulating frequencies, modulation indexes and modulating phases. On the well-known TIMIT database, the ensemble learner classifier is employed for phoneme classification by considering the modulating features.