Keywords : Multiwavenet
Engineering and Technology Journal,
2013, Volume 31, Issue 7, Pages 1418-1430
In this paper, an automatic speaker–independent Arabic word speech recognition system is presented using 3D Radon and Multiwavelet neural network. The approach contains combining multiwavelet theory to neural network which lead to fabricate a Multiwavenet. Position and dilation of the Multiwavenets are fixed and the weights are optimized according to learning algorithm in the network. The feature extraction for real Arabic word signals through 3D radon model is used. The proposed terminology here is training process for some words of all speakers done in Multiwavenet learning phase then test for the other sample speech signals for speakers have been used in Multiwavenet classification phase. Success theory of Multiwavenets has been generalized by extension to biorthogonal wavelets which lead to identification system development. Results show the effectiveness of the proposed system presented in this paper. The accuracy in the detection process was 86% when using utterances outside the training database and around 94% when using the whole utterances database in system test process. The proposed algorithms were implemented using MATLAB2011a.