Lattice Filter Model of Human Vocal Tract

This paper describes the generalized lattice model of human vocal tract for speech analysis, relating it to the all pole type linear prediction. A more natural synthesized speech can be achieved using Auto Regressive (AR) lattice model. Speech analysis is done using linear prediction. The structure of AR lattice model is numerically stable, which is main requirement in speech analysis and synthesis applications. Lattice model uses reflection coefficients; relation of these reflection coefficients with LPC predicted coefficients is given by Levinson Durbin algorithm. Simulation results shows that roots of the polynomial in the denominator of transfer function of lattice model is always guarantee to be inside the unit circle and the stability of the system is achieved.