AI for Oral History Transcription

Oral history records people’s memories and interpretations in their own voices. A transcript makes an interview searchable and can improve access for readers who cannot listen to the recording. Speech-recognition systems can produce a first draft quickly, but oral history includes interruptions, pauses, emotion, code-switching, local names, and words that depend on context. A transcript is an editorial representation, not a neutral copy of the recording. The Oral History Association’s principles emphasize informed consent, transparency, narrator participation, preservation, and clear parameters for access and use. OHA recommends that narrators be offered an opportunity to review the interview and transcript and to approve what is released where possible. Those principles apply whether a transcript is typed by a person or drafted with AI. A narrator’s agreement to an interview does not automatically answer whether audio can be sent to an external transcription service, retained by a vendor, or used to improve a model. Automated transcription may omit a negation, merge speakers, normalize a dialect into standard wording, or replace a person’s name with a familiar but incorrect one. It may remove meaningful pauses or laughter. A human editor should listen to the full audio, check the transcript’s fidelity to the speaker, and mark uncertain passages. Do not silently “correct” grammar in a way that changes voice or meaning. When an interview contains private details, follow the narrator’s permission terms, archive policy, and data-handling agreements before using any external service. A transparent workflow preserves the original audio, transcript versions, correction history, time stamps, speaker labels, model or service used, and access restrictions. Ask the narrator how they want to be identified and whether they approve names, topics, and public release. Provide accessible formats while respecting embargoes or restrictions. AI can make oral history collections easier to search, but ethical stewardship requires consent, context, human verification, and care for the narrator’s words.