Estimation of time-varying AR models of speech through Gauss-Markov modeling

In this work, a new method for estimating the time-varying AR model of speech is presented. Here, the time-varying parameters are modeled as stationary processes. Both the time-varying parameters and their corresponding stationary process are modeled through a common Gauss-Markov model whose state-vector can be estimated through the extended Kalman Filter (EKF) algorithm. The proposed algorithm is different from the earlier methods which use the EKF algorithm. Simulation studies are carried out for both voiced and unvoiced speech. It is shown that the proposed method has less mean-square prediction error than that obtained through the LPC method.