On the performance of Hurst-Vectors for speaker identification systems

Abstract

The performance of Hurst-Vectors (pH feature) for speaker identification systems is presented and discussed in this paper. The pH feature is a vector of Hurst (H) parameters obtained by applying a wavelet-based multi-dimensional estimator (M_dim_wavelets) to the windowed short-time segments of speech. The GMM (Gaussian Mixture Models) and the M-dim.fBm (multi-dimensional fractional Brownian motion) classification systems were considered in the performance analysis. The database - recorded from fixed and cellular phone channelswas uttered by 75 different speakers. The results have shown the superior performance of the M-dim.fBm classifier and that the pH feature aggregates new information on the speaker identity. © Springer-Verlag Berlin Heidelberg 2005.