Maghrebian Accent Recognition Using SVM Classifier and MFCC Features

Abstract

This work aims to design a system to automatically recognize the accent of Maghrebian speakers (Algerian, Tunisian and Moroccan). The recognition system is a support vector machine (SVM) classifier fed by the well-known Mel frequency cepstral coefficients (MFCC) features and their derivatives. The SVM classifier was trained and tested using database of 30 speakers from the three accents. The cumulative sum of the ranked MFCC features seems to have more discrimination between accents than the MFCC features and reaches a classification accuracy of 91.7%.