A modified MFCC feature extraction technique For robust speaker recognition

In Speaker Recognition (SR) system, feature extraction is one of the crucial steps where the particular speaker related information are extracted. The state of the art algorithm for this purpose is Mel Frequency Cepstral Coefficient (MFCC), and its complementary feature, Inverted Mel Frequency Cepstral Coefficient (IMFCC). MFCC is based on mel scale and IMFCC is based on inverted mel (imel) scale. In this paper, another complementary set of features are proposed which is also based on mel-imel scale, and the filtering operation makes these set of features different from MFCC and IMFCC. On the background of this proposed features, the filter banks are placed linearly on the nonlinear scale which makes the features different from the state-of-the-art feature extraction techniques. We call these two features as mMFCC, and mIMFCC. mMFCC is based on mel scale, whereas, mIMFCC is based on imel. mMFCC is compared with MFCC and mIMFCC is compared with IMFCC. The result has been verified on two standard databases YOHO, and POLYCOST using Gaussian Mixture Model (GMM) as the speaker modeling paradigm.