arXiv research

The front-end factor analysis (FEFA), an extension of principal component analysis (PPCA) tailored to be used with Gaussian mixture models (GMMs), is currently the prevalent approach to extract compact utterance-level features (i-vectors) for automatic speaker verification (ASV) systems. Little research has been conduc…

Improves text-dependent speaker verification using neural network supervectors and AUC optimization.

In this paper, we propose a new differentiable neural network alignment mechanism for text-dependent speaker verification which uses alignment models to produce a supervector representation of an utterance. Unlike previous works with similar approaches, we do not extract the embedding of an utterance from the mean redu…

Using supervector fields and graded forms along a morphism, we study the geometry of ordinary differential superequations, extend the formalism of higher order Lagrangian mechanics to the graded context and prove a generalization of Noether's theorem.

The theory of locally anisotropic superspaces (supersymmetric generalizations of various types of Kaluza--Klein, Lagrange and Finsler spaces) is laid down. In this framework we perform the analysis of construction of the supervector bundles provided with nonlinear and distinguished connections and metric structures. Tw…

Given a supervector bundle $E = E_0\oplus E_1 \to M$, we exhibit a parametrization of Quillen superconnections on $E$ by graded connections on the Cartan-Koszul supermanifold $(M;Ω(M))$. The relation between the curvatures of both kind of connections, and their associated Chern classes, is discussed in detail. In parti…

In the spirit of noncommutative geometry we construct all inequivalent vector bundles over the $(2,2)$-dimensional supersphere $S^{2,2}$ by means of global projectors $p$ via equivariant maps. Each projector determines the projective module of finite type of sections of the corresponding `rank 1' supervector bundle ove…