Wavelet packet approximation of critical bands for speaker verification

Abstract

Exploiting the capabilities offered by the plethora of existing wavelets, together with the powerful set of orthonormal bases provided by wavelet packets, we construct a novel wavelet packet-based set of speech features that is optimized for the task of speaker verification. Our approach differs from previous wavelet-based work, primarily in the wavelet-packet tree design that follows the concept of critical bands, as well as in the particular wavelet basis function that has been used. In comparative experiments, we investigate several alternative speech parameterizations with respect to their usefulness for differentiating among human voices. The experimental results confirm that the proposed speech features outperform Mel-Frequency Cepstral Coefficients (MFCC) and previously used wavelet features on the task of speaker verification. A relative reduction of the equal error rate by 15%, 15% and 8% was observed for the proposed speech features, when compared to the wavelet packet features introduced by Farooq and Datta, the MFCC of Slaney, and the subband based cepstral coefficients of Sarikaya et al., respectively.

Speaker recognition

“…It represents the common standard deviation for all Gaussian functions. Moreover, as demonstrated by Specht, (1990), the network is tolerant to the choice of the smoothing factor, and provides robust operation for a relatively wide range of valu (2) The PNN copes well with erroneous training vectors. In many cases a few sparse data samples can be sufficient for optimal performance.…”

“…In validation tests on the NIST 2001 SRE database, whose design and statistical specifications are entirely different from the ones of Polycost, it was confirmed (Ganchev, Siafarikas, Fakotakis, 2004e) that the proposed speech features indeed facilitate the speaker verification process. They demonstrated lower error rates and decision cost than other successful DFT-and DWPT-based speech features, such as the widely used MFCC, and the features of (Sarikaya and Hansen, 2000) and (Farooq and Datta, 2002).Later, in(Siafarikas, Ganchev, Fakotakis, 2005) the same objective criterion was employed in a systematic search for the best wavelet packet tree (among 16 candidates) in a wider area and relaxed margins and bandwidth. The next subsections present the wavelet packet trees design and selection, and the computation of the wavelet packet-based speech features for speaker recognition.…”