Identification of Regional Dialects Using Mel Frequency Cepstral Coefficients (MFCCs) and Neural Network

Dialects may affect the voice recognition because they have an influence on the intonation and the pronunciation of a syllable. Every dialect shows the characteristic of a particular tribe and region. This paper addresses to develop a speech recognition system using Mel Frequency Cepstral Coefficients which are fed to a neural network based on backpropagation algorithm to recognize some particular dialects. The regional dialects used in this paper are Indonesian dialects, namely, Bataknese, Minang, and Javanese. The experiments are conducted to recognize the dialects from training and testing data. The testing data used in the experiments are from different people who have never been trained before so that it is expected that the recognition results will be more valid. The accuracy rate is used to evaluate the performance of the system. The accuracy of dialects recognition for Bataknese, Javanese and Minang are less than 20%, 50% - 80% and 70% - 90%, respectively. This method has succeeded in identifying the regional dialects and the most precise identification is on Minang and Javanese dialects.