Comparison of speech features for Arabic phonemes recognition system based Malay speakers

The selection of the proper feature extraction method is an essential issue for any Automatic Speech Recognition system. This has been conducted in order to choose the suitable feature extraction method for Arabic phoneme recognition system based Malay speakers. In this paper, the implementation of three feature extraction methods involves Frequency Cepstral Coefficients (MFCC), Linear Predictive Coefficients (LPC) and Perceptual Linear Prediction (PLP) has been done. And each feature extraction method is applied on Arabic phonemes in two cases the first case is; the signal is noisy and the second case is the signal is enhanced. The phoneme signals enhancement is achieved using wiener filter. Feed-Forward Neural Network is implemented as a recognizer. The outcome of this study shows that proposed system can give the highest recognition rate with MFCC. The recognition rate is 95.3% and 98.12% in the case of noisy phoneme signal and enhanced phoneme signal respectively. The evaluation and testing the feature extraction methods were based on Arabic phonemes corpus has collected from Malay speakers.