Adaptive ARMA lattice modeling for speech processing

Abstract

An adaptive lattice modeling method for speech processing is proposed. It is a device made of a refined ARMA lattice model. The refined ARMA lattice model is designed as a minimum realization stochastic model. In addition, this paper shows that it is not difficult to determine ARMA orders of the model because of lattice structure. Since the model is identified by using an adaptive approach, the estimation accuracy and the time resolution on the analysis procedure become high enough. Relationship between the refined lattice model and Richards circuits which include loss-less transmission lines is also pointed out. From this association, it is revealed that all ARMA lattice parameters can be transferred to characteristic admittances of transmission lines. Since all admit tanceparameters are physical values, the stability of an estimation model can be checked by observing them directly. From some experimental results, this paper shows that the above properties are suitable for speech processing.