Arabic Phonemes Recognition Using Techniques of Neural Networks

Abstract

The main theme of this research is the recognition of Arabic phonemes using techniques of artificial neural networks, as most of the researches on speech recognition (SR) are based on Hidden Markov Models (HMM). The technique in this research can be divided into three major steps: firstly the preprocessing in which the original speech is transformed into digital form. Two methods for preprocessing have been applied, FIR filter and Normalization. Secondly, the global features of the Arabic speech are then extracted using Cepstral coefficients, with frame size of 512 samples, 170 overlapping, and hamming window. Finally, recognition of Arabic speech using supervised learning method with three types of Neural Networks having completely different strategies is presented. These networks are Multi Layer Perceptron Neural Network MLP, based on Feed Forward Backprobagation. Recurrent Neural Networks which represented with Elman network and finally Kohonen Self Organized Feature Maps with Learning Vector Quantizing network. A comparison between these networks is held. The results of recognition have reached 96.3% for most of the 34 phonemes using Backprobagation Neural Network, using Elman Network it has reached 93.2% and 68% by Learning Vector Quantizing Neural Network. The database used in this research is KAPD (King AbdulAziz Phonetics Database), and the algorithms were written in MATLAB. According to the obtained results, we can confirm that A.N.N can compete the HMM as Arabic language recognizer, which is the main contribution of this thesis.