REFRAMED: Towards Realistic Audio Description Generation for Movies
2026-08-10 • Computation and Language
Computation and LanguageComputer Vision and Pattern Recognition
AI summaryⓘ
The authors talk about Audio Description (AD), which helps visually impaired people understand videos by describing important visuals out loud. Unlike simple captioning, AD needs to be timed carefully to fit between dialogue and only explain what’s necessary. They created a big, detailed dataset called REFRAMED with movie videos, professional AD scripts, subtitles, and screenplays to help computers learn to decide both what to describe and when. They tested current smart systems and found that while these systems do better than basic methods, they still can’t match human experts. This work sets up a new way to improve how machines understand and describe videos.
Audio Descriptionvideo captioningmultimodal learningdatasetscreenplayssubtitlesdialogue gapsmachine learningnarrative contextbenchmark
Authors
Igor Sterner, Mirella Lapata, Alex Lascarides, Frank Keller
Abstract
Audio Description (AD) is a verbal narration of key visual content in videos, enabling access for visually impaired audiences. Unlike standard video captioning, AD is a structured editorial task: descriptions must be inserted into gaps in dialogue and must convey only what is needed to understand the narrative being told. However, existing approaches formulate AD generation in an artificial setting where both the content and timing of descriptions are pre-specified, reducing the task to clip-level captioning. They further rely on noisy transcription and alignment pipelines, and lack the rich parallel data required for modeling narrative context. We introduce a new formulation of AD generation in which models must jointly decide what to describe and when to do it. To support this, we present REFRAMED, a high-quality dataset of 2,023 videos that span 3,302 scenes from 206 movies, with professional AD transcripts (both American and British versions), professional subtitles and aligned screenplays. We also provide a manually curated challenge set that pairs full movies with multiple AD references, together with evaluation protocols that leverage dialogue gaps and multi-reference comparisons. Experiments with state-of-the-art AD systems and multimodal LLMs show that they outperform trivial baselines but fall far short of expert human performance. Our dataset and benchmark establish a new foundation for research on video understanding.