Create a file record first
Record interview ID, date, place, participants, device, file name, duration, consent and sharing restrictions. Make a working copy of the audio. The transcript header should point back to the exact original file.
Draft with speaker labels and timestamps
Identify each speaker consistently and add timestamps at regular intervals or topic changes. Mark inaudible words as [inaudible] and uncertain readings as [word?]. Never fill a gap with the word that merely seems most likely.
Decide how much speech to preserve
A verbatim transcript may include repetitions, false starts and filler words. A reading copy can remove some of those features, but it should not change meaning, grammar in ways that flatten the speaker’s voice, or uncertainty into confidence. Keep both versions when substantial editing occurs.
Separate recollection from editorial fact-checking
Do not rewrite a remembered date inside the quotation because a record differs. Preserve what was said, then add a clearly labeled editor’s note with the source and explanation. This protects both the oral history and the evidence trail.
Use automated transcription as a draft
Speech-to-text can save time, but names, accents, place names and older recordings produce errors. Review while listening, not only by reading. Do not upload private family audio to an AI service unless its retention and training terms fit the consent given.
Create an index and review copy
Add a short summary, people, places, topics and important timestamps. Give the speaker a readable copy and a clear way to request corrections or restrictions. Save the approved version beside the audio, consent note and any later annotations.
Common questions
Clear before you continue.
Should I correct grammar?
For a reading copy, light editing can improve clarity if meaning and voice remain intact. Keep the verbatim version or document the editorial policy.
How do I mark a name I cannot hear?
Use a consistent marker and timestamp, such as [surname unclear, 12:43]. Ask the speaker or another relative rather than silently guessing.
Can AI transcribe the interview?
It can create a draft. Human review, privacy checks and consent are still required, especially for names and sensitive family information.
Sources and further reading
Wiserta summarizes these resources for practical family use. Repository rules and record availability can change; verify details with the holding institution.
A useful next step