Continuous adaptation of linear models with impulsive excitation

Presents a new approach to continuously-adaptive system modelling, designed for the analysis of autoregressive (AR) systems excited by speech signals including an impulsive component. Voiced speech is well represented by such a model, and is used to demonstrate the advantages of the new approach. (1) AR model parameter estimates are more stable in the region of pitch events. (2) A faster adaptation rate can be used, reducing the recovery time after plosives or other sudden changes in signal statistics. The new method is based on multiple simultaneous estimates of each sample, using separate but related estimators. The general concept is illustrated using a linear prediction approach to continuously-adaptive AR modelling, based on the least mean square algorithm.