ISCA Archive Interspeech 2025
ISCA Archive Interspeech 2025

The 2024 NIST Speaker Recognition Evaluation

Craig Greenberg, Lukas Diduch, Audrey Tong, Elliot Singer, Trang Nguyen, Robert Dunn, Lisa Mason, Beth Matys

The 2024 U.S. National Institute of Standards and Technology (NIST) Speaker Recognition Evaluation (SRE) is the latest in a series of SREs conducted by NIST since 1996. The SRE24 evaluation task was automated person detection and had three evaluation tracks: audio, visual, and audio-visual. New SRE24 features included: variable duration enrollment segments, shorter duration test segments, segments containing multiple persons, and updated cost function parameters. 11 teams consisting of 26 sites participated in SRE24. Evaluation results indicate audio-visual fusion produces significant performance improvements over audio or visual systems. Further audio results analyses indicate: use of unrestricted training data produced significant performance gains; some systems performed better using three 10s segments than three 60s segments for enrollment data, with overall differences being less than expected; and, single model systems performed competitively with the fused multi-model systems.