New Front End Based on Multitaper and Gammatone Filters for Robust Speaker Verification

In this paper we present a novel feature extraction algorithm based on multitaper and Gammatone filters for robust speaker verification systems in mismatched noisy conditions encountered in realistic conditions. The idea is to couple the advantage of the low variance multitaper short term spectral estimators with the noise robustness of the auditory Gammatone filterbanks. Experimental results on the TIMIT corpus, with mismatched environment and low signal to noise ratios (SNR) levels, show that the proposed Multitaper Gammatone Cepstral Coefficient (MGCC) features outperform largely the conventional Mel Frequency Cepstral Coefficients (MFCC). Furthermore, and interestingly the MGCC features outperforms at almost all the operating signal to noise ratios the recently proposed Gammatone Frequency Cepstral Coefficient (GFCC) under white, babble and factory noises. This gain in performance is obtained with both the GMM-UBM baseline and the state-of - the art I-vector speaker verification systems.