From recording to quotable transcript
- Add the interview. Drag in the file from your recorder or phone (MP3, M4A, WAV, MP4 and more), or record a video call directly with "Record from App". Scrieb asks what the file is and picks the interview summary.
- Let it transcribe. Whisper and the speaker model run on your computer. Silent stretches are skipped, so the model does not invent text in pauses.
- Check and correct. Rename speakers, fix names in the Document tab. The summary is written from your corrected transcript.
- Export. Word keeps the headings, quotes and bold speaker names; PDF for a fixed copy; plain text or Markdown for analysis software.
How to write and format an interview transcript
Whether you type it yourself or start from software, a transcript others can work with follows a few rules. Set them before the first interview and keep them the same for the whole project.
- Header. Project or story, date and place, interviewer, participant (or a code such as P07), length of the recording, and whether consent was given.
- One paragraph per speaker turn. Start each with the speaker label in bold: Interviewer: and P07:, or real names if the interview is on the record.
- Timestamps. Every few minutes or at each new topic, in the form [00:14:32], so a quote can be found again in the audio.
- Mark what you couldn't hear. [inaudible 00:21:05] or [unclear: Hendricks?], never a guess written as fact. Long pauses as (pause), laughter as (laughs) if it matters for meaning.
- Choose a level of detail. Clean verbatim drops fillers ("um", "you know") and false starts; full verbatim keeps them. Most interviews use clean verbatim.
| Clean verbatim | Full verbatim | |
|---|---|---|
| Fillers and false starts | Removed | Kept: “so, um, we, we started…” |
| Grammar | Left as spoken, not corrected | Left as spoken |
| Pauses and laughter | Only when they change the meaning | Always marked |
| Typical use | Research, journalism, HR, user research | Conversation analysis, legal records |
Example (clean verbatim)
Project: Ward handover study · 14 Oct 2026 · Interviewer: AK · Participant: P07 (nurse) · 52 min · Consent: written
[00:14:32] Interviewer: How does the handover work on a night shift?
P07: We do it at the bedside now. Before, it was in the office, and things got lost between the office and the patient. (pause) The form helps, but only if people fill in the last section.
Interviewer: Who checks that? P07: Nobody, really. [inaudible 00:15:10] the charge nurse, sometimes.
In Scrieb, the speaker labels come from speaker identification: rename "Speaker 1" to Interviewer or P07 once and the whole transcript updates. The Word export puts each speaker's name in bold, so you start from this layout and add your header and markers. Whisper writes close to clean verbatim; for full verbatim, add fillers and pauses back while you listen through.
What makes long interviews hard, and how Scrieb handles it
Length
The recording is written to disk as it happens and the computer is kept awake while transcribing, so a two-hour file does not stall halfway or eat your memory.
Two or more voices
Speaker labels are computed locally. Set the number of speakers when you know it, and run identification again if it miscounted.
Silence and noise
Non-speech parts are skipped. A transcript that looks wrong, such as one phrase repeated over and over, is marked “Check transcript” instead of being summarised.
Language and accents
The language is detected from the speech itself, and you can set it by hand. Large v3 Turbo handles accents and jargon best.
Names and terms
A custom vocabulary spells participants, places and field terms the way you list them.
Video-call interviews
If the other person plays through your speakers, the echo picked up by your microphone is removed after recording.
The interview summary
A transcript of an hour is 8,000 to 10,000 words. The interview summary turns it into something you can work with: an overview, the questions and answers by topic, key quotes word for word with who said them, and open points to follow up. Write your own summary type if your interview guide has a fixed structure.
With a series of interviews, search finds a phrase across every transcript and summary, and the AI chat can answer across all recordings, for example "which participants mentioned the handover form?", naming the interviews it drew on. Both run locally.
Local software vs a transcription service
| Scrieb | Cloud AI service | Human service | |
|---|---|---|---|
| Where the audio goes | Stays on your computer | Uploaded to the vendor | Sent to a typist |
| Cost for 20 one-hour interviews | Free plan or one Pro month | Per-minute or monthly cap | Usually €1 to €2 per minute |
| Turnaround | Minutes, on your machine | Minutes | Days |
| Speaker labels | Yes | Usually | Yes |
| Third party in your data chain | None | Yes | Yes |
A fuller breakdown of the costs: transcription services vs software. Tools compared: best local transcription software.
Who uses it
- Researchers and PhD students: qualitative interviews and focus groups, consistent with consent forms that rule out third parties. More for researchers.
- Journalists: source interviews that never sit on someone else's server.
- Lawyers: client and witness interviews as file notes. More for lawyers.
- HR and user research: candidate interviews and user sessions without personal data leaving the company laptop.
Pricing
Free plan: 1 hour of recording per day with speaker labels and summaries. Pro: no minute limits, one licence for Mac and Windows.
Try it on your own recordings
Frequently asked questions
- Which app transcribes long interviews without terminal setup?
- Scrieb. Install it, drag in a recording of any length, and get a transcript with speaker labels and an interview summary. Whisper runs inside the app, so there is no Python, no command line and no model to download by hand. The recording is written to disk as it happens and the computer stays awake while it transcribes, so two-hour interviews finish.
- How long does a one-hour interview take?
- A few minutes on a Mac with Apple Silicon or a PC with a recent NVIDIA, AMD or Intel GPU; roughly an hour on a laptop without one. It runs in the background, and you can queue several interviews.
- Does it tell interviewer and participant apart?
- Yes. Speaker identification runs on your computer and labels each turn. Rename a speaker once and every occurrence updates. If a two-person interview comes out as three speakers, set the number of speakers and run identification again.
- What does the interview summary contain?
- An overview, the questions and answers grouped by topic, key quotes word for word with who said them, and open points. You can also write your own summary type, for example a structure that matches your interview guide.
- Can I use it for research covered by an ethics approval or consent form?
- Local processing is what most consent forms and data management plans assume: the recording is not shared with a third party. Nothing is uploaded, so there is no processor to list for the content. Check your own protocol, but this removes the usual objection to transcription services.
- What about accents, names and technical terms?
- Choose Whisper Large v3 Turbo for difficult audio, set the spoken language if it is detected wrongly, and add participant names and field terms to the custom vocabulary so they are spelled correctly.
- How do you format an interview transcript?
- Start with a header (project, date, participants, interviewer, consent). Then write one paragraph per speaker turn, with the speaker label in bold at the start and a timestamp every few minutes or at each new topic. Mark unclear words as [inaudible 00:12:40] and long pauses as (pause). Decide up front between verbatim, which keeps fillers like "um" and false starts, and clean verbatim, which removes them, and keep to it across all interviews.
- Verbatim or clean verbatim?
- Clean verbatim for most research, journalism and HR: it reads like speech without the noise. Full verbatim when how something was said matters, for example in discourse or conversation analysis, or for legal records. Whisper produces something close to clean verbatim, so for full verbatim plan a pass to add fillers and pauses back.
- Is it free?
- The free plan includes 1 hour of recording per day with speaker labels, summaries and all export formats, no card needed. Pro removes the limit and covers Mac and Windows with one licence.