Speaker: Sara Barahona Quirós.
Abstract:
In this talk we will present our paritcipation to the NIST 2024 SRE Evaluation in collaboration with Brno University of Technology, Polito, Phonexia, Omilia and CRIM. This evaluation focuses on speaker detection over conversational telephone speech and audio from video, with challenges like cross-source and cross-lingual trials. We will analyze the frontends, backends, scoring methods, and calibration strategies used in both fixed and open tracks, and present the results for the audio-only and audio-visual tracks.