To search, Click below search items.


All Published Papers Search Service


Wavelet-Neural Networks Based Phonetic Recognition System of Arabic Alphabet letters


Moaz Abdulfattah Ahmad, Rasheed M. El Awady


Vol. 10  No. 12  pp. 106-114


This paper is a new trial to recognize the Arabic letters (Arabic Alphabet letters) spoken by any speaker by using artificial neural networks through wavelet technique. This will be useful in converting the spoken words into written text and using of microphone instead of key board, also this can help disable people (handicapped) with limited movement to write any text by voice instead of their hands. A suggested recognition system is implemented to recognize Arabic Alphabet letters of independent speakers. This system is based on analyzing phonetic isolated Arabic alphabet letters. The voice signal is provided to wavelet toolbox to be analyzed. Daubechies (db4) mother function is used in the feature extraction process to produce letters wavelet coefficients. Wavelet coefficients corresponding to each alphabet are used to train multilayer perceptron neural networks to produce recognized binary codes corresponding to each letter. These binary codes (corresponding to alphabet letters) can be decoded to be used as displayed letters on monitors, or printed on paper or used as commands for moving a controlled mechanism. About 95% detection rate has been achieved over large dataset.


Speech processing, Arabic speech recognition, wavelet Transform, ANN, speech technology