Odyssey 2012 - The Speaker and Language Recognition Workshop

June 25-28, 2012

Description and Analysis of the Brno276 System for LRE2011

Niko Brümmer (2), Sandro Cumani (3), Ondřej Glembek (1), Martin Karafiát (1), Pavel Matějka (1), Jan Pešán (1), Oldřich Plchot (1), Mehdi Soufifar (1), Edward de Villiers (2), Jan "Honza" Černocký (1)

(1) Brno University of Technology, Speech@FIT and IT4I Center of Excellence, Czech Republic
(2) AGNITIO, South Africa
(3) Politecnico di Torino, Italy

This paper contains a description of data, systems and fusions developed by the joint team of Brno University of Technology (BUT), Politecnico di Torino (PoliTo) and AGNITIO for the NIST 2011 Language Recognition Evaluation. The primary submission was a fusion of one acoustic and three phonotactic systems, with extensive use of sub-space projections for both approaches. The results are analysed from the view-point of the new NIST measure involving the N = 24 worst language pairs. Some of the results are compared to the MIT-LL submission. As in our previous work, we conclude that having lots of carefully processed data is as important as having good algorithms.

Full Paper

Bibliographic reference.  Brümmer, Niko / Cumani, Sandro / Glembek, Ondřej / Karafiát, Martin / Matějka, Pavel / Pešán, Jan / Plchot, Oldřich / Soufifar, Mehdi / Villiers, Edward de / Černocký, Jan "Honza" (2012): "Description and analysis of the Brno276 system for LRE2011", In Odyssey-2012, 216-223.