Abstarct: A speech signal contains lots of information such as age, gender, emotion and stress, health and accent.One of the challenges in ASR system is varying in accent. It means that if an ASR system train with a special accent and then test it with some different ones the speech recognition ratio degrade drastically. In this thesis we present some novel feature extraction methods for improving the efficiency of a Farsi accent recognition system such as Spectral Centroid Magnitude (SCM) and Spectral Centroid Frequency (SCF) Furthermore we introduce some robust noise feature extraction methods. To classify the results, we use Radial Basis Function (RBF). In addition, we suggest a novel method to improve the efficiency of SVM classifier and a combination of different classifiers. To evaluate the results we use Farsdat data baxse.exprimental result show promising improvement in recognition ratio.