Arabic speech recognition using MFCC feature extraction and ANN classification

This research addresses a challenging issue that is to recognize spoken Arabic letters, that are three letters of hijaiyah that have indentical pronounciation when pronounced by Indonesian speakers but actually has different makhraj in Arabic, the letters are sa, sya and tsa. The research uses Mel-Frequency Cepstral Coefficients (MFCC) based feature extraction and Artificial Neural Network (ANN) classification method. The result shows the proposed method obtain a good accuracy with an average acuracy is 92.42%, with recognition accuracy each letters (sa, sya, and tsa) prespectivly 92.38%, 93.26% and 91.63%.