Classification of Emphatic Consonants and Their Counterparts in Modern Standard Arabic Using Neural Networks

This paper presents the work of acoustic analysis related to Modern Standard Arabic (MSA). The problem of classifting the consonant counterparts in MSA is tackled here. The study considers four phonemes: /d(zeta), partial derivative(zeta)/ and their non-emphatic counterparts /d, partial derivative/ respectively. An accurate automatic classification for those phonemes is to be achieved. Artificial neural networks (ANNs) are used for that purpose. The multilayer perceptron (MLP) is applied to the features extracted from the speech signals. The speech utterances used in this study are from KAPD database. Classification accuracy of 83.9% was achieved.