How to Transcribe an Interview Recording: Steps and Template
Turn an interview recording into a checked transcript: confirm permission, choose a style, label speakers, resolve uncertain wording and verify quotations.
On this page
To transcribe an interview recording, confirm you have permission to transcribe it, decide how literal the transcript must be, run the audio through a transcription tool, then correct it while listening: label the speakers, mark anything you cannot hear and check every quote against the audio. The software produces a first draft. The listening pass is what makes it a transcript you can rely on.
Meetings Brief processing choice: The device-only instructions below require On this device. Live notes also offer Online with Groq when available: selected audio leaves your device in short batches after your processing choice, with shared limits. Online is available for existing recordings: audio goes through private Supabase upload storage to the provider named before upload (Cloudflare Workers AI or Groq), subject to shared usage limits. Check permissions and privacy; AI briefs separately send transcript text to Anthropic.
This guide covers recorded interviews: research interviews, journalism, hiring and user research. If you want notes while the interview is still happening, see the guide to taking notes in live interviews.
Before you start: permission and storage
An interviewee who agreed to be recorded has not automatically agreed to every later use of that recording. Before you transcribe, check three things:
- What the person agreed to. Recording, transcription, quotation and publication can be separate permissions. Research consent forms often say who may hear the audio and how long it is kept.
- Where the audio may go. Some studies, employers and clients restrict which services may process interview material. A tool that uploads the file to a server is a different decision from one that runs on your own computer.
- How the transcript will be stored and shared. Decide where the transcript lives, who can open it and whether names must be removed.
Recording law also varies by place. The guide to recording consent explains the main distinctions, but your organization's policy or an ethics approval may be stricter than the law.
Choose a transcript style before you edit
Decide the standard first, so you don't end up editing the same file twice.
| Style | What it keeps | Use it for |
|---|---|---|
| Verbatim | Words, fillers, repeats and false starts; pauses or other features required by the project | Work whose research protocol or record standard requires that detail |
| Clean verbatim | The speaker's meaningful words, with non-substantive fillers and repeats removed | Readable interview transcripts when the agreed editing policy permits it |
| Separate summary | Your account of the substance, with any checked direct quotes identified | A debrief alongside the transcript; it is not a substitute for one |
Check the project's standard before removing anything. Keep the source audio for the permitted retention period so an editor can revisit a passage. A summary cannot recover the words that were left out.
Step by step: from recording to checked transcript
1. Make a working copy. Keep the original recording untouched. Note the date, the participants and the file name somewhere you will find them later.
2. Choose On this device, then check the language and quality settings. The Meetings Brief transcription tool transcribes in the language you select. It does not detect the language. Its Faster setting uses the Whisper base model and its More accurate setting uses Whisper small. Prepare a short separate excerpt and try both settings before you commit to a long file.
On this device: 3. Transcribe the file. The tool accepts MP3, M4A, WAV, AAC, OGG, OPUS and WEBM audio, and MP4 or MOV video, subject to your browser being able to decode it. It runs in your browser, so the file is not uploaded. The first run downloads the speech engine and model: roughly 96 MB in total on Faster or 260 MB on More accurate. Those totals include the engine, not just the model.
4. Save the raw output. Choose Download .txt before you change anything, and edit a copy in your text editor. Turn on Show the time of each line first if you want the download to include times. Saved results stay in this browser; clearing site data can remove them. The tool does not save your original audio file.
5. Correct while listening. Play the recording and read along. Fix misheard words, especially names, numbers, dates and negatives ("can" and "can't" are easy to swap). The transcript clean-up guide has a full checklist.
6. Label the speakers. The transcription tool does not identify speakers, so add labels yourself during the listening pass (see below).
7. Add timestamps where you will quote. You don't need a time on every line. Put one before each passage you may quote, so you can find it again in seconds.
8. Remove or replace identifying details if required. Follow your consent terms. Replace a name with a code, such as "[Participant 3]", and keep any key separately with restricted access. A job title, small workplace or unusual event can still identify someone after their name is removed. The UK Data Service's participant-protection guidance explains why anonymisation, access controls and participant communication work together.
Label speakers without automatic speaker detection
This tool does not add speaker labels. During the listening pass:
- Use short consistent labels: INT for the interviewer and P1, P2 for participants, or first names if your consent terms allow them.
- Start a new paragraph at each change of speaker.
- When people talk at once, retain the words you can verify and mark the overlap. If you cannot tell who said a phrase, use [speaker uncertain] rather than assigning it to someone.
- If your recording setup saved a separate audio track for each person, transcribe each track separately and merge them by time. Check whether the tracks start together or have different offsets before combining them.
Copy this interview transcript template
Interview transcript
Project / purpose: [study, story or role]
Date and format: [date; in person / video / phone]
Participants: INT = [interviewer]; P1 = [participant or code]
Recording file: [file name and length]
Transcript style: [verbatim / clean verbatim]
Consent notes: [what was agreed about recording, quoting and naming]
Transcribed by / checked by: [names and dates]
[00:00:00] INT: [question]
[00:00:12] P1: [answer]
[00:03:40] P1: [answer containing a passage you may quote]
Unclear passages: [timestamp and what is uncertain]
Worked example: a clean verbatim correction
This example is fictional.
Raw output: "yeah so we we moved the launch to march because the the vendor couldnt um deliver the the sensors in time and lena said she'd chase them"
Clean verbatim after listening:
[00:14:32] P1: Yeah, so we moved the launch to March because the vendor couldn't deliver the sensors in time, and Lena said she'd chase them.
What changed: the repeated words and "um" were removed, punctuation was added, and the listening pass confirmed "couldn't" and the name "Lena". Nothing that changes the meaning was added or removed.
When the audio is hard to hear
Phone-quality audio, crosstalk, echo and strong background noise all reduce accuracy. Don't guess. Mark the problem and move on:
- "[inaudible 00:22:15]" when you cannot make out the words.
- "[unclear: 'forty' or 'fourteen'? 00:22:40]" when you have a candidate but are not sure.
- Return to marked passages at the end with headphones and a slower playback speed.
If a quote you need is inaudible, do not reconstruct it from memory. Ask for clarification or omit it. A later clarification is a new statement, not proof of the original wording. Paraphrase only a meaning you can verify, and keep it outside quotation marks.
Check every quote against the audio
Before a quote leaves your hands, listen to it one more time with the transcript open:
- Are the words exactly what was said, apart from removed fillers?
- Does a cut with an ellipsis change the meaning?
- Is the speaker label right?
- Does the surrounding context support how you are using it?
A transcription model can produce a fluent sentence that the person never said. Compare the quote with the source, including the qualification before or after it; fluent wording is not evidence of accuracy.
Hand over a transcript someone else can check
Before passing it to an editor or researcher, add a short review note:
- Source: the recording filename and version, especially if it has been trimmed.
- Review: who listened through it, when, and which transcript style they applied.
- Open issues: timestamps for uncertain words or speaker labels; do not hide them in an apparently finished document.
- Permitted use: who may read, quote or publish it, and when the recording and transcript must be deleted.
Keep interview evidence separate from your conclusions. For example, "P1 described waiting two weeks" records a participant's account; it does not verify the organisation's actual response time.
Where AI summaries fit
For a recording processed on /transcribe, generating a brief is a separate, optional step. It sends transcript text to Anthropic's AI, and daily limits apply. The brief is designed for meeting summaries and action items; it does not perform a validated research analysis or assess a candidate. Use the checked transcript and your own analysis method for those tasks.
The live notes route behaves differently: a completed capture of at least 150 words requests an AI brief automatically. Disabling automatic email does not disable that text processing. Read the privacy policy and use only a route your consent terms and organisation permit. A manual coding sheet works without an AI summary.
Frequently asked questions
How long does it take to transcribe an interview?
It depends on audio quality, the speakers, the required detail and your device's processing speed. Allow time for a full listening and correction pass as well as the machine draft; do not estimate the job from transcription speed alone.
Can I transcribe a phone interview?
Yes, if you have a recording of it. A web page cannot listen to a call happening on the same phone, so record with the participant's permission using a method your phone or call service supports, then transcribe the file afterward.
Should I keep the recording after transcribing?
Keep it only for the period your consent terms and retention policy permit. Check the rules for the transcript, working copies and identifying key separately; deleting the audio does not automatically permit keeping the text. Delete each when required, and finish any necessary quote checks before its source must be deleted.
Can the tool tell me who said what?
No. The transcription tool produces continuous text without speaker labels. Add labels during your listening pass, or transcribe separate per-person tracks when you have them.
Is a transcript the same as interview notes?
No. A transcript records what was said. Notes record what you concluded. Keep them in separate documents so readers can tell the participant's words from your interpretation.
Sources and scope
Product details and the linked UK Data Service guidance checked on 4 November 2026. The template, review note and example are original working aids; the example is fictional. This guide is not legal advice: consent, transcript standards and retention requirements depend on your jurisdiction, organisation and any research approval.