Speaker verification using various prosodic kernels
Szymon Drgas, Dariusz Zamorski, Adam Dąbrowski · Signal Processing: Algorithms, Architectures, Arrangements, and Applications · 2011
In this article text independent speaker verification problem is considered. The approach in which SVM and prosodic features are employed has been chosen. Prosodic features are modelled by Legendre polynomials. In order to map a sequence of prosodic vectors to fixed dimensional vector three kernels were used: VQ kernel, GMM supervector kernel and Fisher kernel. These three kernels were compared and its combination was evaluated.