Before transcribing an interview, decide which kind of transcript you are producing. That choice, more than the tool, sets how long it will take.
A full verbatim transcript commonly takes 4 to 6 hours per hour of interview. An automatic transcript you proofread brings that down to the proofreading. A selective transcript, limited to the useful passages, is faster still. This guide applies to a dissertation, a thesis, a qualitative study or a professional interview.
The process at a glance
- RecordThe person’s agreement, mic close, quiet room.
- First draftBy hand or with an automatic tool.
- ProofreadAgainst the audio, at reduced speed.
- Apply conventionsSpeakers, timestamps, inaudibles.
- AnonymiseOn a copy, last.
- Archive or deleteThe audio, as announced.
Three levels of transcription
| Level | What it keeps | Best for |
|---|---|---|
| Full verbatim | Everything: hesitations, repetitions, unfinished sentences, laughter, silences. | Discourse analysis, sociology, linguistics, legal use. |
| Clean verbatim | Everything said, without the "um"s, false starts and repetitions. The meaning and the words remain the speaker's. | Qualitative studies, dissertations, HR interviews, journalism. The most common case. |
| Selective | The useful passages word for word, the rest summarised. | Market research, recruitment interviews, user research. |
The same answer, three ways
| Level | What you write |
|---|---|
| Full verbatim | "Well, um… honestly it was… it was bad, you know. I mean, my job had been, um, reorganised [laughs] while I wasn't there." |
| Clean verbatim | "Honestly, it was bad. I mean, my job had been reorganised while I wasn't there." |
| Selective | [00:14:02] P1 describes a difficult return: job reorganised during their absence. Quote: "my job had been reorganised while I wasn't there". |
Which level for a dissertation?
For a semi-structured interview analysed by theme, clean verbatim is the most frequent choice. A sociological or linguistic analysis, interested in how people speak as much as in what they say, needs full verbatim. Either way, ask your supervisor which level is expected before you start: moving from clean to full verbatim means listening to everything again.
Transcripts usually go in an appendix, anonymised, in full or as extracts depending on your programme's rules. The body of the dissertation only quotes the passages that serve the argument.
Before the interview: preparing the recording
- Two devices. A phone backing up a recorder, or the other way round. A lost interview cannot be redone.
- A test on site. Ten seconds, played back, in the interview room.
- The mic near the interviewee. Their voice is what counts; yours, closer to you, will always be audible.
- Consent, recorded. Read a consent sentence at the start of the recording.
"I am going to record our conversation so I can transcribe it. The recording will only be used for [purpose], kept until [date] and then deleted, and your name will not appear anywhere. You can ask to stop at any time. Do you agree?"
Good to know. For video calls, Teams, Zoom and Google Meet offer built-in recording and transcription, depending on your plan. They flag the recording to every participant, but do not replace your consent sentence.
The conventions to set before the first line
- Speakers. A letter or a role, always the same: I for the interviewer, P1, P2 for participants. No first names if the interview must be anonymised.
- Timestamps. At every change of topic, or every five minutes. They let you go back to the audio to check a quote.
- Inaudible. [inaudible 00:12:34] rather than a guess. A guessed word presented as heard is an error nobody catches later.
- Non-verbal. [laughs], [silence], [hesitates], only if your analysis needs them.
- Anonymisation. Replace identifying names, companies and places with a bracketed label: [participant's company], [city].
An extract transcribed with these conventions:
[00:14:02] I: How did going back to work go after your leave?
P1: Honestly, badly. My job had been reorganised while I was away, and nobody had told me. [silence] I found out when I arrived that I no longer handled [client name].
I: Did you raise it with your manager?
P1: Yes, the first week. He said it was [inaudible 00:14:41], that it would be sorted next quarter.
A document template
Every transcript starts with a header. It keeps interviews from getting mixed up and makes your corpus checkable.
| Field | Example |
|---|---|
| Interview code | E07 |
| Date, length, mode | 12 October 2026 · 52 min · in person |
| Interviewer | I (initials in the protocol) |
| Participant | P1 · manager, banking, 8 years in post |
| Level | Clean verbatim |
| Conventions | Timestamp per topic, [inaudible hh:mm:ss], [silence] |
| Anonymisation | Done on 20 October; key table stored separately |
Watch out. The table linking "P1" to the person’s real name is not filed with the transcripts. While it exists, the data is pseudonymised, not anonymous, and is still personal data.
The method, step by step
- Produce a first draft. By hand, or with an automatic tool: see the tools below, and How to transcribe audio to text.
- Proofread against the audio. Playback at 0.75×, text alongside. Fix proper names, numbers, negations, and who said what.
- Apply the conventions. Timestamps, inaudibles, non-verbal.
- Anonymise. Last, on a copy, so you keep a complete working version until the analysis is done.
- Decide what happens to the audio. Keep it as long as announced, then delete it, and write that down in your protocol.
Free tools to transcribe an interview
| Tool | What it does | Where the audio goes |
|---|---|---|
| noScribe | Automatic transcription built for research interviews, speaker separation, a correction editor. Needs a recent computer. | Your computer |
| Whisper | The open-source speech-recognition model many tools are built on. | Your computer |
| Sonal | Free software from the University of Tours to transcribe, categorise and analyse interviews. | Your computer |
| Word (Transcribe) | Transcript split by speaker, 300 min a month with Microsoft 365. | Microsoft servers |
| Teams, Zoom, Meet | Transcription of interviews held by video call, depending on the plan. | The vendor's servers |
For a dissertation, tools that run on your computer avoid the question of handing data to a third party. Your institution sometimes has an approved solution too.
Group interviews and focus groups
With several people, the hard part is knowing who speaks. Do a round of introductions at the start of the recording, each person saying their first name or code. Note the seating order on paper. During the discussion, name the person you hand the floor to. And plan twice as much proofreading time as for a one-to-one interview.
The time it takes
For a dissertation of ten one-hour interviews, the gap is several weeks of work. An automatic first draft does not excuse proofreading: it changes the nature of the work, from typing to checking.
Interviews and GDPR
A recorded interview holds personal data: the voice, and everything the person says. Four rules apply, whether you are a researcher, a recruiter or a consultant:
- Inform. The person knows why you record, who will access the recording and how long it will be kept.
- Get their agreement to record before you start. In France, recording someone without their knowledge is a criminal offence, as we explain in Recording a work meeting in France: what the law says.
- Limit retention. Audio is not kept forever. Once the transcript is validated and anonymised, it has often served its purpose.
- Choose your tools. Uploading an interview to an online service hands the person's data to a third party. For academic work, your institution's data protection officer can tell you what is allowed.
Quoting an interview
A quote taken from a clean verbatim transcript may be smoothed, never rewritten. If you cut, mark it with [...]. If you add a word for clarity, put it in brackets. And keep the timestamp in your notes: it is what lets you stand behind the quote if the person disputes it.
After transcribing: analysis
Once interviews are transcribed, thematic analysis means finding and labelling the passages that deal with the same topic. A word processor and highlighters are enough for a handful of interviews. Beyond that, qualitative analysis software does the job: NVivo, ATLAS.ti and MAXQDA are paid, Taguette and Sonal are free.
Frequently asked questions
How long does it take to transcribe an interview?
By hand, 4 to 6 hours per hour of interview in full verbatim, 3 to 5 hours in clean verbatim. With an automatic first draft, 1 to 2 hours of proofreading, depending on recording quality.
Should I transcribe the interviewer's questions?
Yes. An answer only makes sense with the question that prompted it, and how the question was asked is part of what your reader must be able to judge.
Should I correct spoken language?
In clean verbatim, you restore written forms ("going to" rather than "gonna") and remove hesitations, without touching vocabulary or how ideas are built. In full verbatim, you correct nothing.
Can I use AI to transcribe interviews for a dissertation?
Yes, provided you proofread the whole result against the audio, have informed the person, and pick a tool that fits your confidentiality commitments. A tool that runs on your computer settles that last point.
Where do transcripts go in a dissertation?
In an appendix, anonymised, in full or as extracts depending on your programme's rules. The body of the text only quotes the useful passages.
Key takeaways
- Pick the level first: full verbatim, clean or selective.
- Set conventions before the first line: speakers, timestamps, inaudible.
- By hand, plan 4 to 6 h per hour of interview; an automatic first draft reduces the work to proofreading.
- Keep the key table separate: without that, nothing is anonymous.
- Inform, get agreement, and do not keep the audio without a reason.
Notus records the interview and returns the transcript to your phone, processed and hosted in France, ready to proofread.
Discover Notus →