Pair-of-sequences SVM speaker verification

We present a new concept of speaker verification based on a target independent decision system. The basic principle is to build a system that decides whether two sequences were pronounced by the same speaker. In our view, this system is aimed to complement traditional ones. While the principle is quite general, in this paper we use an SVM scheme to implement it. To do so, we conceive a kernel between pair of sequences using GMM distributions estimated on each given sequence. We present experiments on NIST Speaker Recognition Evaluation. The individual performance of the new system is similar to the GLDS-SVM, and the fusion of both outperforms the baseline GMM system.