Abstract: This paper describes the two systems submitted by the UVigo-GTM group for the Albayzin 2010 language verification evaluation. They were initially thought to perform speaker recognition and verification, so they use language-independent information to apply the algorithms to language verification treating the problem as a pattern recognition task. The principal system consists in a dimensionality reduction approach that transforms the data into a lower dimensionality subspace by performing a two-stage process to reduce the dimensionality and extract a discriminative subspace. The alternative system uses the Non-negative Matrix Factorization to obtain a representation of the data in terms of a set of basis functions, obtaining the utterances represented as a feature vector of lower dimensionality.
Index Terms: fishervoices, language identification.