New Feature Vector based on GFCC for Language Recognition

Here a new form of feature vectorbased on Gammatone Frequency Cepstral Coefficients (GMFF) for language recognition is proposed. The major battle neck in degradation of language recognition(LR) performance is the presence of noise and mismatchedenvironment present in the speech signal.For any language recognition, the default feature vectors are MFCC, but the performance degrade sin the presence of noise and mismatch conditions. From the literature, it is observed that GFCC has very good robustness against additive noise. In this work, a new feature vector using GFCC is introduced for language recognition tasks. The new feature vector based on GFCC for the GMM LR system task showed superior performance when it is compared tothe conventional MFCC feature vector-based GMM LR system.