Wavelet packet based speaker verification

In an attempt to find out a more appropriate representation of a speech signal for the task of speaker recognition, we study alternative ways to represent speakers’ voices individuality. A novel wavelet packet based set of speech features, apposite for speaker recognition, is proposed. We exploit the capabilities offered by the plethora of existing wavelets, along with the powerful set of orthonormal bases provided by wavelet packets that allow an effective manipulation of the frequency subbands. Our scheme differs from previous wavelet-based works, primarily in the wavelet-packet tree design which follows the concept of critical bandwidth, as well as in the particular wavelet basis function that has been used. Our baseline text-independent speaker verification system, which has participated in the 2002 NIST Speaker Recognition Evaluation, was used as a platform to study the practical significance of the proposed speech parameters. Comparative experimental results confirm the assertion that the proposed speech features outperform MFCC, as well as previously used wavelet features, on the task of speaker verification.