Efficient Code Excited Linear Predictor Using Redundant Vector Quantiser Representations

The description of a novel linear prediction (LP) model-based coding technique was presented, where the advantages of multiple non-orthogonal domain representations of LP coefficients and the residuals were exploited in conjunction with vector quantisation. The LP model-based coding was applied to speech signals and the resulting improvement in performance was demonstrated in terms of the reconstruction quality. Sample results using speech signals exhibited improved reconstruction quality which was achieved at the cost of a moderate increase in the computation associated with the representation of the signals in multiple domains