University of the basque country (gtts@ehu) system for the nist 2017 language recognition evaluation

This paper briefly describes the language recognition systems developed by the Software Technology Working Group (http://gtts.ehu.es) of the University of the Basque Country (EHU) for the NIST 2017 Language Recognition Evaluation. The submitted system uses the the Brno University of Technology (BUT) 80 dimension bottleneck features [1] trained on FisherEnglish (2423 triphones) and follows the Total Variability Factor Analysis (i-Vector) approach [2]. The i-Vector extractor (1024 Gaussians and 400 dimensional i-Vector) is based on the Sidekit Toolkit [3] and it is followed by a Gaussian Linear Classifier and a Discriminative Gaussian Backend. Linear logistic regression calibration is applied to the final scores using the FoCal Toolkit [4].