Improving Access to Oral Histories: ASR for Accented and Variant English in Libraries and Archives

Abstract

Libraries and archives care for long-form audio collections such as oral histories, interviews, lectures, and community recordings, yet these materials often remain difficult to search, caption, describe, and share. This poster presents an ongoing project to develop an automatic speech recognition (ASR) workflow that improves access to spoken collections, especially those featuring accented English, regional varieties of English, mixed-language speech, and community-based vocabulary. Designed with libraries and archives in mind, the system creates speaker-attributed, searchable transcripts that can support accessibility, discovery, description, and research use. It is especially suited to long recordings that are common in oral history and archival work. The project also explores fairer ways to evaluate transcription quality when recordings include local terms, variant spellings, or language-specific form. This custom ASR program is intended to support staff, improve access, and strengthen stewardship of cultural memory held in long-form audiovisual collections for diverse communities.