Same-Language Subtitles for Low-resource Languages: A Case of Bundelkhandi

Abstract

Same-language subtitles enhance consumers’ experience for audiovisual content for both hearing impaired population. However, while high-resource languages can benefit from automatic subtitling, subtitles are seldom available for content creators in regional languages. This limits audience engagement on their content, which often is independently produced. This paper presents Project Saurakhi, a platform for generating same-language subtitles in regional languages. To achieve this, we first extract community-generated YouTube videos serve as the primary data source for this project. The current dataset comprises 63 hours of Bundelkhandi speech sourced from 207 YouTube videos across 19 content creators. And second, the technical workflow integrates automated stages with manual refinement via a mobile annotation platform. As regional language content grows both in independent productions, and in over-the-top platforms, Project Saurakhi aims to train women participants in rural India to become proficient in providing subtitles in their native languages.

Resources

Details