Adaptive estimation of time-varying model order in the ARMA speech analysis

A technique for estimating the numbers of poles and zeros of the vocal tract transfer function in speech analysis is proposed based on the simultaneous estimation of autoregressive moving average parameters (SEARMA) method. The problem is solved using statistical decision theory as well as physical properties of the speech production process. The detection of zeros can be controlled by the choice of an appropriate cost function, and the interframe correlation can be taken into account by introducing the a priori probability. The validity of the proposed method is demonstrated by analysis of both synthetic and natural speech sounds.<
>