{
  "id": 17982,
  "url": "https://arxiv.org/abs/2608.01310",
  "title": "FATE: Frame-Level Audio-Visual Temporal Embedding",
  "summary": "When a dog opens its mouth and barks, humans naturally recognize what the sound is and when it occurs. Building audio-visual models with this same ability requires representations that capture both semantic and temporal alignment. Current approaches fall short on one side or the other: embedding models match semantic but lose temporal information; synchronization models capture temporal offsets but lack semantic understanding. To bridge this gap, we propose FATE, Frame-level Audio-visual Tempora",
  "authors": "Kaisi Guan, Bingzi Zhang, Xihua Wang, Ying Ba, Xin Cheng, Yijing Chen",
  "category": "research",
  "topics": "safety-alignment",
  "orgs": null,
  "regions": null,
  "published_at": "2026-08-01T20:00:00.000Z",
  "fetched_at": "2026-08-11T05:10:37.351Z",
  "source_slug": "hf-daily",
  "source_name": "HuggingFace Daily Papers",
  "source_homepage": "https://huggingface.co/papers",
  "ethics_ai_record_url": "https://ethics.ai/record/17982",
  "original_url": "https://arxiv.org/abs/2608.01310",
  "evidence_status": "source-only",
  "attribution": "via ethics.ai"
}