On the use of wavelet and Fourier transforms for speaker verification

In this paper, we investigate the use of the wavelet transform for text-dependent speaker verification task. An extensive experimental framework have been conducted using several approaches: (a) a fixed structure wavelet transform, (b) an adaptive wavelets scheme according to a given criterion and finally (c) a best basis according to the best wavelets packet transform. A speaker dependent library tree has been built, corresponding to the best structure for a given speaker. Finally, (d) a fusion of information combining wavelets and Fourier. Experiments have been conducted using 60 speakers, extracted from Yoho Database and a phone-based HMM speaker verification system. Results have been compared with MFCC features.