Comprehensive evaluation of theoretical approximations for spectral quantization performance

In this work, recently derived theoretical approximations for high rate vector quantization (VQ) are reformulated to cleanly separate the effects of the VQ codevector density and the local Voronoi region shapes on overall VQ performance. Numerical evaluation of the resulting theoretical expressions is performed, which allows comparison of the relative importance of codevector density and Voronoi region shapes for a variety of different conditions. In particular, results are presented which compare root mean squared (RMS) vs. mean squared (MS) optimal quantizers, full band vs. partial band log spectral distortion (LSD) quantizers, LPC vs. K vs. LAR vs. ASIN vs. LSP vs. cepstral coefficients, 0th order vs. 1st order recursion, and optimal vs. weighted mean squared error (WMSE) vs. mean squared error (MSE) quantizers.