An AR Model Identification Method using a Prior Information and Its Application to a Spectral Estimation of Speech Signal

An Extension of the Burg Method Based on the Principle of the Minimum Cross Entropy

Abstract

A novel Auto Regressive (AR) model parameter estimation method is proposed, which can utilize a prior information as well as time series data, by extending the Burg method on the basis of the Minimum Cross Entropy (MCE) principle. As a practical application of the proposed method, we consider an approach to spectral estimation of speech data. In general, effectiveness of a prior information to spectral estimation results depends on the variation of speech signal. Thus we introduce an algorithm to determine the usage of a prior information, based on the divergence measure defined by the Kullback information. Finally, the estimation results for real speech data illustrate improved performance in comparison to the Burg method.