In 2012 NIST held the latest in an ongoing series of text-independent speaker recognition evaluations (SREfs). The 2012 NIST Speaker Recognition Evaluation (SRE12) was the largest and most complex SRE to date, including over 100 million trials. Several aspects of SRE12 were new; most significantly, NIST released in advance of the evaluation target speaker training data from six preexisting corpora, and systems were permitted to utilize joint information from target speakers for speaker modeling and test segment scoring. Results from the evaluation suggest that systems found it easier to reject non-target trials where the test speaker was among the target speakers.
Bibliographic reference. Greenberg, Craig S. / Stanford, Vincent M. / Martin, Alvin F. / Yadagiri, Meghana / Doddington, George R. / Godfrey, John J. / Hernandez-Cordero, Jaime (2013): "The 2012 NIST speaker recognition evaluation", In INTERSPEECH-2013, 1971-1975.