15th Annual Conference of the International Speech Communication Association

September 14-18, 2014

Summary and Initial Results of the 2013-2014 Speaker Recognition i-Vector Machine Learning Challenge

Désiré Bansé (1), George R. Doddington (1), Daniel Garcia-Romero (2), John J. Godfrey (2), Craig S. Greenberg (1), Alvin F. Martin (1), Alan McCree (2), Mark Przybocki (1), Douglas A. Reynolds (3)

(2) Johns Hopkins University, USA
(3) MIT Lincoln Laboratory, USA

During late-2013 through early-2014 NIST coordinated a special i-vector challenge based on data used in previous NIST Speaker Recognition Evaluations (SREs). Unlike evaluations in the SRE series, the i-vector challenge was run entirely online and used fixed-length feature vectors projected into a low-dimensional space (i-vectors) rather than audio recordings. These changes made the challenge more readily accessible, especially to participants from outside the audio processing field. Compared to the 2012 SRE, the i-vector challenge saw an increase in the number of participants by nearly a factor of two, and a two orders of magnitude increase in the number of systems submitted for evaluation. Initial results indicate the leading system achieved an approximate 37% improvement relative to the baseline system.

Full Paper

Bibliographic reference.  Bansé, Désiré / Doddington, George R. / Garcia-Romero, Daniel / Godfrey, John J. / Greenberg, Craig S. / Martin, Alvin F. / McCree, Alan / Przybocki, Mark / Reynolds, Douglas A. (2014): "Summary and initial results of the 2013-2014 speaker recognition i-vector machine learning challenge", In INTERSPEECH-2014, 368-372.