Parametric Representation of Voiced Speech Phoneme using Multicomponent AM Signal Model

In this paper, the amplitude modulated (AM) sinusoidal signal model is used for parametric modelling of speech signal. The multicomponent AM sinusoidal signal model is found to be well suited for voiced speech phonemes. The Fourier-Bessel (FB) expansion is used to separate individual components of a multicomponent signal. Parameter estimation of the AM sinusoidal signal model is presented in detail. The cyclo-stationary analysis is used for frequency estimation of the AM sinusoidal signal with high accuracy under noisy signal condition. The symmetric Itakura-Saito (IS-COSH) distance measure is used for comparison of original and reconstructed signals.