Java and Sunda dialect recognition from Indonesian speech using GMM and I-Vector

Abstract

Recognition of dialect is a field of research in the field of language recognition. The purpose of this study is to compare the best modelling features and techniques for recognizing Javanese and Sundanese dialects in Indonesian speech using MFCC and combination of MFCC and pitch feature candidates and comparing GMM and I-Vector modelling techniques. This research uses machine learning approach for training process and model testing. As a result, the use of the combination of MFCC and pitch feature and the I-Vector modelling technique has the advantage to recognize Java and Sunda dialect.