Multi-scale fractal dimension for speaker identification system

Abstract

This paper looks at the extraction of nonlinear turbulence information of the speech signal using Multi-Scale Fractal Dimension (MFD) as the additional feature to the convectional MFCC features. The MFD is estimated using both Box-Counting and Minkowiski-Bouligand dimensions. The proposed framework uses this feature extraction together with sub-band based speaker modeling rather than the wide-band approach. Results showed that the proposed framework with Box-Counting feature extraction improves the performance of the classical wideband approach with up 10% identification rate.