Généralisation du noyau GLDS pour la Vérification du locuteur par SVM
Jérôme LOURADOUR, Khalid DAOUDI, Francis BACH.

The Generalized Linear discriminant Sequence (GLDS) kernel provides good performance in SVM speaker verification in NIST SRE (Speaker Recognition Evaluation) evaluations. It is based on an explicit mapping of each sequence to a single vector in a feature space using polynomial expansions. Because of practical limitations, these expansions has to be of degree less or equal to 3. In this paper, we generalize the GLDS kernel to allow not only any polynomial degree but also any expansion (possibly infinite dimensional) that defines a Mercer kernel (such as the RBF kernel). We conceive a new kernel, and makes it tractable using a method of data reduction adapted to kernel methods : the Incomplete Cholesky Decomposition (ICD). We present experiments on NIST SRE database, that show good perspective for our new approach.