MEMORY NOTCH BLOG

How to transcribe qualitative research interviews

Transcribe an interview for qualitative research in five steps: record, machine-transcribe, check against audio, anonymise, code. Plus where the audio may go.

Published November 6, 2026 · By Cho Yin Yong

To transcribe an interview for qualitative research, record it with consent and transcribe it with a tool your ethics approval names. Then correct the draft against the recording, replace names with pseudonyms, and import the text into your analysis software to code it (tag passages by theme). Decide two things first: how literal to be (every um, or cleaned up), and where the audio is allowed to go.

Where the audio goes is the decision that stalls projects. Several universities now say in writing which transcription services their researchers may use, and a cloud service your approval doesn't name can mean amending it. The University of Bath updated its research data guide on November 4, 2026. It says “you must have a data sharing agreement in place before you transfer any personal data to third parties.”

Record: a separate file per interview, named by participant number and date.
Transcribe: a machine pass gives you a draft, not a transcript.
Check: listen through and fix every line against the audio.
Anonymise: names and places become labels in square brackets.
Code: import the plain-text file into your analysis software and tag passages by theme.

How to transcribe an interview for qualitative research, step by step

Five steps take a recording to a codable transcript, and the third one, checking against the audio, is the one you can't skip.

Record. Get consent on the recording itself, or on the signed form your ethics approval describes. Use one file per interview. Name it with the participant's code and the date, not their name. File names travel further than transcripts do.
Machine transcript. Run the audio through the tool your approval names. An on-device tool transcribes while you talk, so the draft exists when the interview ends. A cloud service needs the upload first.
Check against the audio. Play the recording and read along, a section at a time. Fix misheard words and names, mark gaps as [inaudible], and confirm who said each line. Machine transcripts slip on fast speech, accents, slang and poor audio, as one university guide warns.
Anonymise. Replace people, places and organisations with codes in square brackets. Sheffield's transcription proforma uses Name1 and Name2 for people, and Place1, Hosp1 or Org1 for the rest. Each code gets a timestamp, so you can find the original moment later. Keep the key that maps codes to real names in a separate, protected file.
Code. Import the checked, anonymised text into your analysis software and start coding.

Check first, then anonymise. Once a name has become [Name1], you can no longer hear whether the recognizer got it right.

Verbatim vs intelligent verbatim: which should you use?

Use full verbatim when your analysis looks at how people speak, and intelligent (clean) verbatim when it looks at what they say. Sheffield's proforma draws the line in one sentence each. Verbatim means “do not shorten words, include all ‘err’, ‘like’, ‘you know’ etc.” Intelligent verbatim means “exclude err/um/erm”.

	Full verbatim	Intelligent (clean) verbatim
Fillers (err, um, erm)	Kept	Removed
“Like”, “you know”	Kept	Your template decides
Words as spoken	Not shortened	Your template decides
Keeps evidence of	How things are said	What is said
Checking time	Longer: every filler has to be right	Shorter

If you study hesitation, repair or turn-taking, those are your data, and a clean transcript deletes them. For thematic work, fillers mostly add reading time. Ask your supervisor which style your method expects. Write the choice into your transcription template, and use it for every transcript in the study. If you need full verbatim, listen for fillers during the check, because a machine draft won't always keep every one.

The proforma is worth reading wherever you study. It asks whether to include the consent exchange at the start, interruptions from people who walk into the room, and the chat about paperwork at the end. Settle those once, before transcript one.

Why do ethics boards care where the interview audio goes?

Ethics boards care because a recording of a research participant is personal data, and every service that touches it is somewhere that data now lives. Their guidance asks you to name the service, keep the data where your agreements cover it, and tell participants.

Three universities show how specific these rules have become.

University of Canterbury. Its AI transcription guidance, dated September 24, 2026, advises against AI meeting tools for “in confidence” or sensitive research data. Those tools store and process data outside New Zealand. Researchers should try the university's licensed call transcription first, because it stays on university servers, or else the university's own offline service.

If an AI service is used at Canterbury, the participant information sheet must name it. The recording and transcript must be deleted from the service within 24 hours of upload.

University of Auckland. The transcription page approves the university's own video-call tools when you're signed in to a university account. It also approves two open-source transcribers on a university-managed virtual desktop. It names several popular cloud services as not approved, and says to “be aware of the need for service agreements and approved data transfer options.”

University of Bath. The guide notes that many online services store data on servers in other countries, and that a data sharing agreement must be in place before personal data goes to a third party.

Where the audio goes	What your approval usually needs
A human transcription service	A data sharing or confidentiality agreement with the transcriber
A cloud AI transcription service	The service named in the protocol and the consent form, plus its data location and retention
Your institution's own tools	Usually pre-approved; check the account and storage it uses
An app that transcribes on your own computer	The device, its storage and its encryption described in your data plan

Checked November 6, 2026.

On-device transcription removes the third party from the transcription step. It doesn't remove your obligations. The laptop is now where the data lives, so your data management plan has to cover it: disk encryption, backups, and when the audio is deleted. Your board may still prefer its own managed tools, and it decides. Ask before the first interview, not after the tenth. The longer argument for keeping recordings off servers is in why your meeting notes don't need the cloud.

How to record in-person and remote interviews on a Mac

For an in-person interview, the laptop's microphone records the room. For a video call, record the microphone and the computer's system audio together, so your voice and the participant's both land in one file.

In person. Put the laptop between you and the participant, on a solid table, away from fans and coffee machines. Do a ten-second test and play it back before you start. A phone works too, and lets the laptop stay closed: the guide to recording an in-person meeting covers placement and phone apps.

Remote, on a video call. The participant's voice arrives as system audio, the same sound you hear. On macOS, an app needs the Screen Recording permission to capture it, even when no video is saved. Wear headphones. They cut the duplicate audio you get when the speakers feed back into the microphone. Some platforms also offer their own recording and transcript; if your university's guidance points you there, as Canterbury's and Auckland's do, that is the approved route.

Speaker labels. Transcription apps that separate voices label them by number, not by name: Speaker 1, Speaker 2. In a two-person interview, work out which number is you in the first minute, then replace the labels with your own codes, such as Interviewer and Participant 7, using find-and-replace. Labels are most reliable when people take turns. When you both talk at once, check those lines by ear. What speaker diarization is explains how the separation works and where it fails.

How to format and cite an interview transcript

The citation style of the American Psychological Association (APA) does not treat your own participants' words as citations. The APA page on quoting research participants says these quotations are part of your original research, so they get no reference list entry and are not personal communications.

What the page does ask for:

Quotations under 40 words go in quotation marks in the text; 40 words or more go in an indented block quotation.
Say in the text that the quotations come from participants.
Follow the confidentiality and anonymity terms agreed during consent, for example by using pseudonyms and obscuring identifying details.

APA's quoting page covers quoting participants in your paper, not the layout of the transcript file. For the layout, use your department's template, or a published one. Sheffield's proforma includes a template with numbered lines and a running header, so you can point to a passage by line. Swansea University's library sends its students to the same page. If your supervisor or journal asks for something different, theirs is the rule that applies.

Which tools fit each step?

Different tools cover different steps, and none of them covers all five. A recorder makes the draft, an editor helps you check it, and analysis software does the coding.

Tool	What it does	Where the audio goes
oTranscribe	A free, open-source web app for typing a transcript while you play the recording, with timestamps, keyboard playback controls and Markdown or plain-text export	Its site says “your audio file and transcript never leave your computer”
Your institution's Teams or Zoom	A recording and a machine transcript of a call	The university's own account and storage
ATLAS.ti	Coding and analysis; imports transcripts in VTT, SRT, TXT, DOCX and RTF, and has its own automatic transcription service	Check ATLAS.ti's terms for its transcription service
NVivo	Coding and analysis; imports Word, rich text and plain-text documents, and M4A audio	Your project file
Memory Notch	Records and transcribes on a Mac, iPhone or Android phone, with numbered speaker labels	The device that recorded it

Checked November 6, 2026.

oTranscribe is a good fit for the checking step if you transcribe by hand or want a clean editor beside the audio. It doesn't transcribe automatically. You type, and it keeps the audio under your fingers.

Where Memory Notch fits

Memory Notch is a recorder and transcriber. On a Mac, it records your microphone and system audio from the notch at the top of the screen, writes a live transcript with numbered speaker labels, and saves the audio, a plain-text transcript and a Markdown note to ~/memory. Recording and transcription run on the computer, offline. The app makes three network requests, for a revoked-licence list, update checks you can turn off and update downloads, and the privacy policy says none of them carries a recording.

For the checking step, use a transcript timecode to jump back to a moment in the audio, or copy the transcript with timecodes from More actions. Enhanced transcript re-transcribes a saved recording more carefully. The Vocabulary tab takes the names and jargon of your field, with what they get misheard as. It improves the odds, and it doesn't guarantee spelling.

What it doesn't do:

No coding or tagging. Bring the .txt file into your analysis software.
No automatic summaries.
Speakers are numbered, not named, and overlapping voices are hard to separate.
The transcript is a draft. Check every quote by ear.
Phones can't record other apps' audio. Use them for in-person interviews only.

The Mac app needs Apple Silicon and macOS 26 or later. Prices are in American dollars: $79 once for up to five of your own Macs, with a year of updates, or $7.99 a month, cancel anytime. The free trial records up to five minutes per recording and saves everything captured. The iPhone and iPad app and the Android app are free to download, and a one-time $5.99 in-app purchase removes the limit; each is bought separately and keeps its own recordings.

Recording on your own device doesn't replace your participants' consent or your ethics approval. It only decides where the audio sits once they've agreed.

Frequently asked questions

How long does it take to transcribe a one-hour interview?

Checking time depends on the audio quality, the accents, how many people talk at once and your transcription style. Checking a machine draft is usually quicker than typing from scratch. Time your first check, then plan the rest of the project from that number rather than from a rule of thumb.

Should I transcribe my own interviews?

Transcribing or checking your own interviews means you hear every hesitation again, which is often where analysis starts. The Bath guide also lists anonymising as you go as an advantage of self-transcription. The cost is time. A machine draft moves most of that time from typing to checking.

Can I use AI to transcribe research interviews?

Often yes, if your ethics approval names the tool. Some universities restrict cloud AI transcription for sensitive data, require participants to be told the service's name, or limit how long data may stay on it. Check your institution's guidance, and update your protocol before switching tools mid-study.

Is verbatim or clean verbatim better for thematic analysis?

For thematic analysis, intelligent (clean) verbatim keeps what people said and drops fillers like “um”, which rarely carry a theme. If your method looks at how people speak, such as hesitations or interruptions, use full verbatim. Check with your supervisor, then use one style for every transcript.

Do I cite interview transcripts in APA?

Not in the reference list. APA Style says quotations from your own research participants are part of your original research, so they get no reference entry and aren't treated as personal communications. Say in the text that they come from participants, and use pseudonyms where your consent terms require them.

How do I record a Teams interview on a Mac?

Use an app that captures the microphone and system audio together, and grant it screen-recording permission in System Settings so it can hear the call. Wear headphones to reduce duplicate audio. If your university approves the call app's own recording and transcription, that may be the route your ethics approval expects.

Sources

Download Memory Notch · More from the blog