Language Identification System using MFCC and SDC Features

Abstract

Speech recognition is useful to recognize spoken words and phrases from speech samples and to converts them into a machine-readable format. It has classes as text recognition, speaker recognition and language identification. The speech signal is basically intended to carry information about the linguistic message, it also contains the language specific information. In this regard, this work undertakes the study and implementation of Language Identification System using Gaussian Mixture Model (GMM) classifiers. Language identification system finds specific language from speech samples. This system is based on Mel-frequency Cepstral Coefficients (MFCC), Shifted Delta Cepstral Coefficients (SDC) and MFCC-delta coefficients feature extraction techniques. MFCC gives the information about human vocal tract shape. SDC and MFCC-delta gives the information about phonemes and temporal information of language. The proposed system applied on 8 Indian languages as Konkani, Assamese, Bengali, Hindi, Oriya, Malayalam, Tamil and Panjabi. Experiments are done to compare two combination viz MFCC attached with SDC and MFCC-delta attached with SDC. Comparison shows that language identification system using MFCC-Delta and SDC feature extraction gives better results with 51.38% accuracy on 8 Indian languages.