Speaker ethnic identification for continuous speech in Malay language using pitch and MFCC

Voice recognition has evolved exponentially over the years. The purpose of voice recognition or sometimes called speaker identification, is to identify the person who is speaking. This can be done by extracting features of speech that differ between individuals due to physiology (shape and size of the mouth and throat) and also behavioral patterns (pitch, accent and style of speaking). This paper explains an approach of voice recognition to identify the ethnicity of Malaysian people. Pitch and 13 Mel-Frequency Cepstrum Coefficients (MFCCs) are extracted from 52 recorded continuous speech in Malay for use as features to train the classifiers using Tree, Naïve Bayes, Nearest Neighbors and Support Vector Machine (SVM) and another 10 recorded speeches are used for testing. The results reveal that the use of a combination of pitch and 13 coefficients for features extraction and training the data using SVM provide better accuracy (57.7%) than the use of only 13 coefficients (53.8%).