Discrimination of Different Serbian Pronunciations from Shtokavian Dialect

This paper proposes a new methodology for discrimination of different pronunciations in the Shtokavian dialect of the Serbian language. At the first, the written language (Unicode text) is converted into codes according to the energy status of each character in the text-line. Such a set of codes is seen as a grayscale image. Then, the local structures of the image are explored by local binary operators. It creates a vector set which differentiates various pronunciations of the Serbian language. The experiment is performed on fifty documents given in Serbian language. A comparison performed between the proposed method and the n-gram method shows its clear advantage.