A HIGH-PERFORMANCE HTK BASED LANGUAGE IDENTIFICATION SYSTEM FOR VARIOUS INDIAN CLASSICAL LANGUAGES

Abstract

Language identification is one of the major research areas in the field of speech processing and tremendous works has been done on that. One of the great bottleneck of language identification system is, for languages which are having closely related pronunciation, it is very difficult to classify them. In our experiments, we have considered seven Indian classical languages and Mel frequency cepstrum with their delta cepstral feature are utilized as features and HMM is used as a classifier. Performance of the system is analysed using HTK based continuous density HMM with MFCC features. The state of the art HTK with MFCC features produced 98.1% with 3 Gaussian mixtures and 100% accuracy with 10 Gaussian mixtures even with small amount of training samples. The same method can be utilized for other languages also.