Arabic isolated word recognition using general regression neural network

Author

Amrouche, Abderrahmane ; Rouvaen, Jean Michel

Author_Institution

Lab. for Speech Com. & Signal Proc., USTHB, Bab Ezzouar

Volume

2

fYear

2003

fDate

30-30 Dec. 2003

Firstpage

689

Abstract

In this paper the results of the general regression neural network (GRNN) applied to Arabic isolated word recognition are presented. The architecture proposed consists in two parts: a pre-processing phase which consists in segmental normalization and feature extraction and a classification phase which uses neural networks based on nonparametric density estimation. In order to accomplish such comparison the GRNN and the traditional multilayer perceptron (MLP) have been tested. The results obtained by using a large set of Arabic digits shows that the neural networks based on the general regression improve the recognition rate more than those based on the feed forward back propagation error

Keywords

feature extraction; feedforward neural nets; multilayer perceptrons; neural nets; speech recognition; Arabic digits; Arabic isolated word recognition; classification phase; feature extraction; feed forward back propagation error; general regression neural network; multilayer perceptron; nonparametric density estimation; recognition rate; segmental normalization; Automatic speech recognition; Computer networks; Feature extraction; Hidden Markov models; Natural languages; Neural networks; Phase estimation; Random variables; Speech recognition; Yield estimation;

fLanguage

English

Publisher

ieee

Conference_Titel

Circuits and Systems, 2003 IEEE 46th Midwest Symposium on

Conference_Location

Cairo

ISSN

1548-3746

Print_ISBN

0-7803-8294-3

Type

conf

DOI

10.1109/MWSCAS.2003.1562380

Filename

1562380