Speech recognition techniques using RBF networks

Author

Phillips, William J. ; Tosuner, Caner ; Robertson, William

Author_Institution

Dept. of Appl. Math., Tech. Univ. Nova Scotia, Halifax, NS, Canada

Volume

1

fYear

1995

fDate

15-16 May 1995

Firstpage

185

Abstract

This paper presents a pattern recognition approach, based on whole word patterns, to speaker independent automatic speech recognition of isolated digits. We use the decomposition of the spoken word into subacoustic words to ensure time alignment of the significant portions of the input´s acoustic characteristics and those of the reference patterns. The Isodata clustering algorithm is used by the radial basis function (RBF) network to create reference templates and classification of the speech samples

Keywords

feedforward neural nets; pattern recognition; speech recognition; Isodata clustering algorithm; RBF networks; acoustic characteristics; isolated digits; pattern recognition; radial basis function network; reference patterns; reference templates; speaker independent automatic speech recognition; speech recognition techniques; speech samples classification; spoken word decomposition; subacoustic words; time alignment; whole word patterns; Artificial neural networks; Automatic speech recognition; Clustering algorithms; Feature extraction; Logic testing; Loudspeakers; Pattern recognition; Radial basis function networks; Signal processing algorithms; Speech recognition;

fLanguage

English

Publisher

ieee

Conference_Titel

WESCANEX 95. Communications, Power, and Computing. Conference Proceedings., IEEE

Conference_Location

Winnipeg, Man.

Print_ISBN

0-7803-2725-X

Type

conf

DOI

10.1109/WESCAN.1995.493968

Filename

493968