LAUSR

A new multicomponent multitone amplitude and frequency-modulated signal model for parametric modelling of speech phoneme (voiced and unvoiced) is presented in this paper. As the speech signal is a multicomponent non-stationary signal, the Fourier–Bessel expansion is used to separate all individual components from the multicomponent speech signal. The parameter estimation is done by analysing the amplitude envelope (AE) and instantaneous frequency (IF) of the signal component separately. The AE and IF functions for separated components are extracted by using the discrete energy separation algorithm. The amplitude-modulated signal parameters and the amplitude of the signal are estimated by analysing the AE function, whereas the frequency-modulated signal parameters and the carrier frequency of the signal are estimated by analysing the IF function. This technique is found to be quite efficient for accurate parameter estimation of the speech phoneme. As an illustration of model-based speech processing, the proposed model is used for various speech signal processing applications.