NarrAD: Automatic Generation of Audio Descriptions for Movies with Rich Narrative Context

Abstract

This paper addresses accessibility for visually impaired individuals by automating narration generation for long-form videos. The framework reflects the narrative context of the entire movie, including the storyline, names of characters and places. It leverages movie scripts while ensuring audio descriptions don't overlap with dialogue. NarrAD achieves the highest user experience and movie comprehension in user studies with 600 subjects, outperforming prior approaches on the MAD dataset.