Speaker-independent speech recognition by means of functional-link neural networks

Author

de Arriaga, F. ; El Alami, M.

Volume

2

fYear

2000

fDate

2000

Firstpage

1018

Abstract

The purpose of the paper is to show the improvements so far obtained in the solution of the speech recognition problem by means of functional-link neural networks. The specific problem deals with speaker independent recognition of Spanish vowels, a problem of theoretical and practical complexity. The voice Fourier spectrum has been obtained and from it the mel-cepstrum and cepstrum coefficients, which have been used as inputs for the neural architecture in charge of the recognition problem. Different polynomial functions such as Newton and Lagrange developments are used as functional expansions of the network. The results so far obtained show improvements in the recognition rate in all cases with respect to those obtained with the multilayer perceptron

Keywords

Newton method; learning (artificial intelligence); polynomials; speech recognition; Spanish vowels; cepstrum coefficients; functional-link neural networks; mel-cepstrum coefficients; polynomial functions; speaker-independent speech recognition; voice Fourier spectrum; Cepstrum; Character recognition; Fourier transforms; Frequency domain analysis; Lagrangian functions; Multilayer perceptrons; Neural networks; Polynomials; Speech recognition; Stability;

fLanguage

English

Publisher

ieee

Conference_Titel

Pattern Recognition, 2000. Proceedings. 15th International Conference on

Conference_Location

Barcelona

ISSN

1051-4651

Print_ISBN

0-7695-0750-6

Type

conf

DOI

10.1109/ICPR.2000.906247

Filename

906247