Time-varying singular value decomposition: an application on speech-like signals

Author

Bosch, L. F M Ten

Author_Institution

Dept. of Language & Speech, Nijmegen Univ., Netherlands

fYear

1991

fDate

14-17 Apr 1991

Firstpage

3197

Abstract

A combination of two methods for signal parameter extraction is presented. One method involves singular wave decomposition, while the other deals with time-varying predictive coding. The combination makes it possible to find a prediction vector as well as its first derivative from a short speech segment. Time-varying algorithms will be of interest in the study of signals whose correlation changes smoothly with time, such as voiced speech sounds in the vicinity of plosive bursts. Their applicability for signal regions with fast spectral changes is demonstrated

Keywords

eigenvalues and eigenfunctions; encoding; filtering and prediction theory; speech analysis and processing; SVD; fast spectral changes; numerical experiments; plosive bursts; signal parameter extraction; singular wave decomposition; speech-like signals; time-varying predictive coding; time-varying singular value decomposition; Frequency; Linear predictive coding; Matrices; Matrix decomposition; Natural languages; Predictive coding; Signal analysis; Singular value decomposition; Speech; Tin;

fLanguage

English

Publisher

ieee

Conference_Titel

Acoustics, Speech, and Signal Processing, 1991. ICASSP-91., 1991 International Conference on

Conference_Location

Toronto, Ont.

ISSN

1520-6149

Print_ISBN

0-7803-0003-3

Type

conf

DOI

10.1109/ICASSP.1991.150135

Filename

150135