How to Transcribe an Oral History Interview Without Losing Your Mind

Why Transcription Is Worth Doing (Even If It's Slow)

Audio is not searchable. When a family member needs to find the moment your grandmother mentioned a specific name — a town, a church, a cousin who left and never came back — they need text, not a playback bar to scrub through blind. A transcript is quotable in a genealogy document, shareable with relatives who weren't in the room, and attachable to a family history file. It is what makes an oral history submission-ready for an institutional archive: most historical societies and tribal collections will not accept raw audio alone. They need a formatted transcript with documentation before they can process a donation. The recording you spent an afternoon capturing is significantly more useful — and significantly more permanent — with a transcript alongside it. That is the case for doing this. It is a real time investment. Everything below is about making it as efficient as possible.

Before You Start: The 3-Minute Setup

Three things before you type a single word. They take about three minutes total and will save hours over the course of a transcript.

Put on headphones. Not speakers. Headphones let you hear clearly at low volumes, reduce listening fatigue, and catch details a speaker setup misses — consonants at the ends of words, names spoken softly, a voice that drops at the beginning of a hard answer. If you've been playing the recording back on laptop speakers, you have been missing things. Headphones on before you start.

Slow the playback to 70–75% speed. Every transcription app has this setting — it is not just for accessibility. At 75% speed, words that blur together at normal pace become distinct. Regional speech, soft delivery, and fast talkers all become clearer. This is the single biggest productivity unlock in transcription work. Set it before you begin and leave it there for the entire session.

Set a chunk size. Ten minutes of audio per session. Transcription fatigue is real — after 45 minutes of focused listening and typing, accuracy drops and the corrections start to compound. Working in 10-minute chunks means each session is finite. You finish a chunk, you stop, and the work is done for today. The recording will wait. This is not a shortcut; it is how you actually finish the transcript instead of abandoning it halfway through.

Your Options: Free Tools vs. AI Auto-Transcription

There are three approaches. Each has a clear use case.

Manual transcription with oTranscribe (free). oTranscribe.com is a free browser-based tool that puts your audio player and your text editor side by side in the same window. It has keyboard shortcuts to pause, rewind five seconds, and adjust speed — all without leaving the typing field. It auto-saves to the browser, supports foot-pedal input, and exports clean text. It is the best free option for high-quality recordings, interviews with heavy dialect or regional vocabulary, or any situation where accuracy is critical. If you plan to donate the recording to an archive, this is the tool to use for the final review pass.

AI auto-transcription (fast but imperfect). Otter.ai offers a free tier with 300 minutes of transcription per month — enough for two or three interviews before hitting the limit. Rev.com charges $1.50 per minute with turnaround under 24 hours, which works out to roughly $90 for a one-hour interview. Apple Dictation can be used in a pinch for short clips by playing audio near the device microphone, though it is inconsistent and unreliable for anything longer than a few minutes.

The honest limitation: AI transcription misses names, accents, crosstalk, and quiet voices. It will confidently produce a wrong word and give you no indication that it guessed. Family names, place names, and any term outside its training data will come out wrong. Auto-transcripts always need a review pass. An unreviewed AI transcript is not a finished document.

The hybrid method (recommended). Run the recording through AI first to get a rough draft, then listen through the audio once at 75% speed while reading and correcting the text. This is 60 to 70 percent faster than typing from scratch, and the correction pass typically surfaces things you missed on the first listen anyway — an emphasis, a pause before a difficult answer, a name the AI rendered phonetically wrong. For most people doing this for the first time, this is the right workflow: fast start, careful review, finished document.

Formatting That Makes a Transcript Actually Useful

A transcript that no one can navigate is nearly as useless as no transcript. Four formatting decisions make the difference.

Label speakers every time they change. Not just at the start of the document — every turn. Use a consistent format: all caps, colon, space (RUTH: or INTERVIEWER:). When a new voice begins, start a new line. This makes the transcript scannable, searchable, and clear to anyone who wasn't in the room.

Use brackets for unclear audio. [unclear] means you heard something but could not confirm the word. [inaudible] means you could not make it out at all. Never guess at a word you cannot hear — a guessed word that enters the record becomes fact by the time the document is three years old and no one remembers which words were heard clearly and which were invented. If you have a strong sense of what was said, use [sounds like: Hargrove?] to flag it as an interpretation, not a confirmed transcription.

Include timestamps every 2–3 minutes. The format is simple: [12:34]. Timestamps let you go directly back to a passage in the audio when you need to verify something, and they let future researchers — or family members who want to hear a specific moment — do the same. They take a few seconds to add and save real time later.

Note significant non-verbal content. [long pause], [emotional], [laughs]. These are not distractions — they are data about how something was said, which often carries as much weight as the words themselves. A ten-second pause before an elder answers a question about land loss is part of the answer. A transcript that erases that pause is a less honest record than the recording it came from.

Handling the Hard Parts: Accents, Crosstalk, and Emotional Content

Three situations come up in almost every family oral history transcription. Here is how to handle each.

Accents and regional speech. Do not correct dialect. An elder who says “fixin' to” or drops final consonants or uses a construction that sounds wrong to a standardized ear is producing a historical record, not a grammar error. Transcribe what was said, phonetically if needed, then add a bracketed note if you think clarification would help a future reader: [phonetic] or a brief gloss in brackets. The dialect is part of the evidence. Standardizing it is an edit you don't have the right to make — it removes information from the record.

Crosstalk. When two people talk at once, mark it with [crosstalk] and capture what you can from each voice. Do not construct a clean exchange that wasn't there. If you can isolate one speaker, transcribe what they said and mark the other [inaudible]. If you can't isolate either, the whole exchange gets [crosstalk — both speakers, inaudible]. The record is incomplete; it should say so.

Painful or heavy content. Some family oral histories contain testimony about violence, loss, or experiences the narrator has never spoken aloud before. That content belongs in the transcript exactly as it was said — it is part of the historical record. But you are not just a transcriptionist. You are the person who knew this person, or knew of them, or came to sit with them for a reason. It is legitimate to stop. Close the laptop. Come back in a day. The document will wait, and you will do more accurate, more careful work when you are not trying to hold the weight of the story and the precision of the text at the same time.

What to Do With the Transcript After

Once the transcript is finished, three things complete the file.

Save as both .docx and .pdf. The .docx stays editable — you will want to correct it as new information comes in, and family members who add context will need to make changes. The PDF is the archival copy: fixed, shareable, printable, and compatible with any system. Both formats should live in the same folder as the audio file.

Name it with the same convention as the audio. [Name] — [Date] — [Topic]. For example: Grandma Mae — June 2026 — Migration Story. Whatever naming convention you used for the audio file, match it exactly for the transcript. This keeps the two files connected through any file system, across any device, and for any person who wasn't part of your organizational decisions when you set the folder up.

Store them together. The transcript, the audio file, and the signed consent form in the same folder is what institutional archives require. A transcript without a corresponding audio file is harder to verify. An audio file without a consent form cannot be donated. When all three are present, named consistently, and stored together, you have a donation-ready package — a complete record that a historical society, tribal archive, or HBCU special collection can accept and process. That is one step from a permanent home. The guide on donating oral history to historical societies and tribal archives covers what comes next.

For more on what happens after the transcript is done — how to organize it, how to use it as a research document, and how to share it with the rest of the family — see the guide on what to do with oral history after you record it.