A GMM supervector approach for spoken Indian language identification for mismatch utterance length

Gaussian mixture model-universal background model (GMM UBM) supervectors are used to identify spoken Indian languages. The calculated from short-time MFCC, its first and sec derivatives. UBM builds a generalized language model, mean adaptation transforms it duration normalized language-specific GMM. Multi-class support vector machine artificial neural network classifiers labels the supervectors. Experimental evaluations performed using 30 speech utterances nine languages comprised five Indo-Aryan four Dravidian languages, extracted all India radio broadcast news data-set. Eight smaller data-sets were manually derived study effect of training test mismatch. In mismatch conditions, identification accuracy decreases with decrease in train utterance duration. Investigations showed that 32-mixture ANN classifier has optimal performance.