arxivcs.SDcs.AI2026-07-05
Auto-AEG: Scalable Data Construction for Open-Vocabulary Audio Event Grounding
Zihan Zhang, Xize Cheng, Wenhao Yan, Tong Zhang, Dongjie Fu, Boyun Zhang, et al.
Large Audio-Language Models (LALMs) reason fluently about sound yet struggle to localize precisely when events occur, while classical Sound Event Detection attains frame-level precision only over a closed label set. At the intersection of these paradigms lies the task of Open-Voc…