A novel approach to variable dimension vector quantization of harmonic magnitudes

Harmonic coders explore the harmonic structure of a given signal to represent it efficiently, and have become quite successful recently where several standardized speech coders are based upon it. Quantization of harmonic magnitudes is an integral component of harmonic coders. Since the harmonic magnitude sequences can be considered as vectors of variable dimension, the technique of variable dimension vector quantization (VDVQ) is highly suitable for their representation. In this paper a novel implementation of VDVQ is described; experimental data show the superiority of the technique which is compared to existent methods.