arxiv
PublishedJune 18, 2026 at 4:00 AM
—neutral
Revisiting Active Speaker Detection: An In-the-Wild Benchmark for Generalization and Robustness
Publisher summary· verbatim
arXiv:2505.21954v2 Announce Type: replace-cross Abstract: We present UniTalk, a novel dataset emphasizing challenging scenarios to enhance model generalization for the task of active speaker detection (ASD). Previously established benchmarks such as AVA predominantly comprise old movies and thus exh
Stay posted· Newsletter
A 5-min weekly brief — top movers, price watch, story of the week.
Discussion
No replies yet. Be first.
Related coverage
The Bubble Brief
WEEKLYRead AI insights every Tuesday — top movers, new releases, story of the week.
Originally published on arxiv ↗