Vector quantization of articulatory shapes.

A key element in the eventual application of the articulatory approach to low bit-rate coding is the ability to derive model parameters from the speech signal. Focusing on the vocal tract model, the aim of the work reported herein is 1) to take a step toward accurate representation of the cross sectional area function of the vocal tract, and 2) to propose a procedure amenable to a coding scenario. Therefore, we describe a new technique for quantizing the articulatory space that can be used to meet our stated objectives.