Robust Non-recursive AR Analysis of Speech Signal

Abstract

This part considers non-recursive algorithms, also known as algorithms with block (packet) processing, for estimation of parameters of AR model of speech signal. The basis of these algorithms is gathering of measurement data at an interval of a particular length, based on which the unknown parameters of the adopted AR model are estimated. It is started from the assumption that the signal at the considered interval is stationary, i.e., one can select segments of speech signal on which the system for speech production can be modeled by a stationary (time-invariant) model.