A two-mode structure of adaptive linear predictor

Abstract

In this paper, we investigate a two-mode structure of adaptive linear predictor with application to speech coding. Initially, this structure has independently adapting low-order cascaded stages that use forward-backward linear prediction algorithm. Later, it switches to a regular transversal structure using least mean square (LMS) algorithm. This method enjoys fast convergence rate at the beginning and accurate tracking of inputs in the steady state.