Large Margin GMM for discriminative speaker verification

Gaussian mixture models (GMM), trained using the generative criterion of maximum likelihood estimation, have been the most popular approach in speaker recognition during the last decades. This approach is also widely used in many other classification tasks and applications. Generative learning in not however the optimal way to address classification problems. In this paper we first present a new algorithm for discriminative learning of diagonal GMM under a large margin criterion. This algorithm has the major advantage of being highly efficient, which allow fast discriminative GMM training using large scale databases. We then evaluate its performances on a full NIST speaker verification task using NIST-SRE’2006 data. In particular, we use the popular Symmetrical Factor Analysis (SFA) for session variability compensation. The results show that our system outperforms the state-of-theart approaches of GMM-SFA and the SVM-based one, GSL-NAP. Relative reductions of the Equal Error Rate of about 9.33% and 14.88% are respectively achieved over these systems.