Title :
Normalization and polygon error detection for split VQ of line spectral frequencies
Author :
Smith, Aaron M. ; Ashley, James P. ; Jasiuk, M.A. ; Weiming Peng
Author_Institution :
Labs., Motorola Inc., Schaumburg, IL, USA
Abstract :
A technique for improving the performance of split vector quantization (VQ) of line spectral frequency (LSF) parameters is presented, wherein a normalization process is used to make all combinations of the codebook entries valid for preserving the LSF ordering property. An undesirable consequence of normalization is a reduction in the ability to detect bit errors in the LSF codebook indices by exploiting the ordering property. A new method, polygon error detection (PED), is presented which shows promise for overcoming this problem. The use of PED is also shown to be beneficial when normalization is not performed
Keywords :
error detection; spectral analysis; speech coding; vector quantisation; LSF codebook indices; bit error detection; codebook; line spectral frequencies; line spectral frequency parameters; normalization; ordering property; performance; polygon error detection; speech coder; split VQ; split vector quantization; Data mining; Decoding; Equations; Frequency; Histograms; Natural languages; Parameter extraction; Speech processing; Training data;
Conference_Titel :
Speech Coding, 2000. Proceedings. 2000 IEEE Workshop on
Conference_Location :
Delavan, WI
Print_ISBN :
0-7803-6416-3
DOI :
10.1109/SCFT.2000.878422