You Described, We Archived: A Rich Audio Description Dataset

Area of Science:

• Computer Science
• Human-Computer Interaction
• Linguistics

Background:

• The You Described, We Archived (YuWA) dataset is a unique collection of audio description (AD) data.
• It was compiled through a collaboration between San Francisco State University and The Smith-Kettlewell Eye Research Institute.
• Data spans from 2013-2022, sourced globally via the YouDescribe platform.

Purpose of the Study:

• To present the You Described, We Archived (YuWA) dataset for research applications.
• To highlight the utility of crowd-sourced audio descriptions for various AI and computational linguistics tasks.
• To provide a rich resource for studying video understanding and accessibility.

Main Methods:

• The YuWA dataset comprises audio description data collected via the YouDescribe platform.
• YouDescribe facilitates user-generated audio descriptions for YouTube videos, involving volunteer describers and requests from blind and visually impaired (BVI) users.
• Data includes audio description tracks, metadata on describers and viewers, and collection timelines.

Main Results:

• The YouDescribe platform has engaged over 3,000 volunteer describers, creating more than 5,500 audio-described videos.
• The YuWA dataset contains worldwide audio description data from 2013-2022.
• The dataset is publicly available at https://youdescribe.org/.

Conclusions:

• The YuWA dataset offers significant potential for research in artificial intelligence, machine learning, and natural language processing.
• It serves as a valuable resource for advancing audio description technologies and video-language grounding.
• The dataset supports interdisciplinary research spanning computer science, accessibility, and sociolinguistics.