Title :
KLT-based classified VQ with companding
Author :
Kim, Moo Young ; Kleijn, W. Bastiaan
Author_Institution :
TMH Speech Signal Process. Group, R. Inst. of Technol., Stockholm, Sweden
Abstract :
In this paper, we introduce companding in our Karhunen-Loeve transform (KLT)-based classified vector quantizer (CVQ). The companding utilizes the shape advantage of a vector quantizer (VQ) more efficiently, and the Voronoi-region shape of code vectors approaches a hyper-sphere as the number of classes increases. With companding, even the single-class KLT-CVQ outperforms a conventional code-excited linear predictive (CELP) coding, both in terms of SNR and codebook search complexity.
Keywords :
Karhunen-Loeve transforms; computational geometry; speech coding; transform coding; vector quantisation; CVQ; KLT-based classified VQ; Karhunen-Loeve transform-based classified vector quantizer; Voronoi-region shape; code vectors; codebook search complexity; companding; hyper-sphere; shape advantage; Covariance matrix; Density functional theory; Distortion measurement; Eigenvalues and eigenfunctions; Performance analysis; Shape; Signal processing; Speech processing; Testing; Vectors;
Conference_Titel :
Speech Coding, 2002, IEEE Workshop Proceedings.
Print_ISBN :
0-7803-7549-1
DOI :
10.1109/SCW.2002.1215706
