Who it's for

For researchers. Your ethics protocol said the audio stays local.

Qualitative interviews, focus groups, fieldwork in places with no coverage. The data management plan you wrote to get approval usually contains a sentence about where recordings live, and most transcription services contradict it on the first upload.

Use it forInterviews, focus groups, fieldwork with no coverage, and your own analytic memos.
The argumentNo third party receives the audio, so the sentence in your data management plan stays true after transcription too.
Not forTeam projects. Nothing syncs, there is no shared library, and eight people coding together need an institutional tool.

When you'd use it
The risk
Why Just Transcribe
The rules
Setting up
Where it fails
Answers
Next

Rows of white archive boxes on wooden shelves. — Photo: Luke Caunt / Unsplash

Five points in a project

The work where transcription time is the real bottleneck, not the analysis.

Semi-structured interviews

Forty interviews of an hour each is the standard shape of a qualitative study and roughly two hundred hours of transcription if you do it by hand. Paid services cost real money per hour and introduce the party your consent form did not mention.

Machine transcription with human correction is now the normal compromise, and the only question is where the machine is.

In the app: record, transcribe on the device with a large model, then correct in the editor with find and replace. Export as text with timestamps for your coding software.

Focus groups

Six voices, overlapping, in a room with hard surfaces. The hardest audio in qualitative research and the place where speaker attribution matters most, because who said it is part of the data.

On-device speaker separation gets you a draft; the correction is still yours.

In the app: save each participant's voice at the start of the session, transcribe with speaker separation, then rename speakers to pseudonyms in the editor before exporting.

Fieldwork without coverage

Rural sites, other countries, boats, mountains, anywhere the mobile network is theoretical. Cloud transcription means carrying a backlog home and uploading forty hours over a hotel network, which is both slow and the worst possible moment to be transferring participant data.

In the app: transcribe the same evening, on the device, with no connection. You leave the field with text rather than with a queue.

Multilingual research

Interviews in a language you read slowly, or not at all. Having the original transcribed in its own language, then machine translated for triage, tells you which interviews need a professional translator's time.

The translation is a reading aid. It does not go in the thesis.

In the app: fifty-nine transcription languages with automatic detection, more than 1,600 through Omnilingual, and on-device translation of a finished transcript into fifty-five. Keep the original transcript as the data of record.

Your own field notes and memos

The observation made walking away from the site is the one that turns into an analytic memo six months later, and it is the one most reliably lost.

Spoken memos are faster, longer and more honest than written ones, and they become searchable text.

In the app: record immediately, transcribe later, and search the full text of every memo across the project. Folders keep interviews, memos and admin apart.

The gap between the protocol and the workflow

Almost every qualitative protocol contains a version of the same promise: recordings will be stored securely, access will be limited to the research team, and data will be deleted after a stated period. Participants are shown that promise and consent on the strength of it.

Then the interviews get transcribed, because transcribing by hand takes five hours per hour of audio, and the fastest route is a web service. At that moment the promise becomes untrue, usually without anybody noticing, because nobody thinks of a transcription tool as a place where data is stored.

What an upload does to your protocol

It adds a party your participants never consented to, in a country your ethics application probably did not mention, with a retention period you did not set. If the service is outside your jurisdiction it may also create an international transfer you have not assessed.

It is also the kind of thing that is discovered late, during a data protection review or a funder audit, at which point the honest answer is that the consent form does not match what happened.

What Just Transcribe lets you write

That audio is recorded and transcribed on a single encrypted device, that no third party receives it, and that no copy exists outside the research team's control. Those sentences are checkable and they survive a review.

The App Store privacy label reads Data Not Collected. The app connects twice: to download a speech model you choose, and for the purchase.

This does not write your data management plan, pass your ethics review, or anonymise anything. It makes one common sentence in a protocol true instead of aspirational.

Why Just Transcribe

The feature list is ordinary; the reason to pick it is what is missing from the chain.

No third party, so the consent form stays true. Nothing is transferred and nobody outside the team receives the audio.
No account. Nothing ties a participant's voice to a login, a subscription or an institutional email address.
Works in the field, permanently offline. Transcription, translation and summarising all complete with the radios off.
Speaker separation with local voice profiles. Focus group transcripts arrive attributed, and the profiles never leave the device.
Exports your analysis software can read. Text with speaker labels and timestamps, plus PDF, SRT and VTT.
Editing built in. Find and replace, speaker renaming, correction. Pseudonymise before the transcript ever leaves the device.
EUR 9.99 once. No per-minute pricing, which matters when a project has two hundred hours of audio and a grant that has already been spent.

No third party receives the recording, so the sentence in your data management plan about where the audio lives stays true after transcription as well as before it. Transcription is the step where that promise is most easily broken without anyone noticing.

What to write in the protocol

Accurate descriptions you can use, and the places where this app does not help you.

Wording is yours and your ethics committee's. This is what is technically accurate, not what is sufficient.
Where audio is recorded and processed	On a single mobile device. The speech model runs locally and no audio or text is transmitted to the developer or any third party.
Third party processors	None for transcription. The developer operates no server that receives study content, and the App Store privacy label reads Data Not Collected.
International transfers	None arising from transcription, because there is no transfer. Transfers arising from your own storage, backup or collaboration remain yours to assess.
Special category data	If your interviews touch health, beliefs, sexuality, ethnicity or political opinion, GDPR Article 9 applies to the recording. Keeping it on one device is a safeguard, not a lawful basis.
Anonymisation	Not provided. The app transcribes what was said, including names. Pseudonymisation is a manual step you do in the editor before export.
Retention and destruction	Manual. You delete files on the device on the schedule you promised. There is no central retention policy doing it for you.

Do not copy these into a protocol without your own checking. An accurate description of the tool is not the same as an adequate data management plan.

Two sentences for the data management plan

Audio is recorded and transcribed on a single encrypted mobile device. The speech model runs locally and no audio or transcript is transmitted to the developer or any third party, so no processor relationship and no international transfer arise from transcription.

Check it against your own setup before you paste it in. An accurate description of the tool is not an adequate data management plan by itself, and it says nothing about your backups, which is the part reviewers ask about.

Setting it up before fieldwork

Four steps, ideally before the ethics submission rather than after it.

Describe the workflow in the protocol accurately. Including that transcription happens on the device, that no third party receives audio, and who on the team has access to the device. Ethics committees respond well to specificity.
Download the largest model your phone can hold, before you leave. For most languages that is Large v3 Turbo, around 954 MB; for a language nothing else covers, the largest Omnilingual build the phone offers, up to around 7 GB. Accuracy matters more than speed when you are transcribing overnight rather than in a queue, and in the field there is no second chance to download it.
Agree a pseudonymisation and export convention with the team. Rename speakers in the editor, replace identifying names with find and replace, and export only the cleaned transcript. Decide the file naming once.
Decide the backup before you need it. There is no server, so device loss is total loss. An encrypted backup to an institutional store, done deliberately and described in the protocol, is the responsible version of that.

The awkward one: a phone holding forty participant interviews is now a research device, not a personal one. Encryption, passcode, and a clear answer to what happens if it is stolen all belong in the data management plan rather than in your head.

Where Just Transcribe is the wrong tool for a project

One of these is a problem for team research and you should weigh it before committing.

No team access, at all. Nothing syncs and there is no shared library. On a four-person project, every transcript moves between people as an exported file. For some teams that alone rules it out.
No anonymisation. Names are transcribed as spoken. Pseudonymising is manual work in the editor, and forgetting it is the most likely way this bites you.
No CAQDAS integration. Nothing writes into NVivo, ATLAS.ti, MAXQDA or Dedoose. You export text and import it there.
No verbatim conventions. It does not mark pauses, overlaps, laughter or intonation. Conversation analysis needs a different transcription practice and probably a human.
No backup. No server means device loss is data loss unless you arranged an encrypted backup yourself.
No Windows version. It runs on iPhone, iPad, Mac and Android, and the Mac app needs macOS 26 or later, so a team on Windows laptops or older Macs is a team with a workflow problem.

For a solo project, a PhD or fieldwork, those limits are mostly tolerable. For a distributed team of eight coding collaboratively, they are not, and an institutional tool with a proper agreement is the better answer.

Straight answers

Will my ethics committee accept on-device transcription?

In my experience they like it, because it makes the data flow shorter and easier to describe, but they decide and not me. What helps is describing it precisely: the model runs on the device, no audio or text is transmitted, the developer operates no server that receives study content, and the App Store privacy label reads Data Not Collected. Specific and checkable beats reassuring.

How accurate is it compared with a professional transcription service?

Below a good human transcriptionist and broadly comparable to other machine transcription when you use one of the larger models on clear audio. Expect to correct names, technical vocabulary and anything said over the top of someone else. The realistic workflow is machine draft plus human correction, which is what most researchers already do with paid services.

Can I use it for a focus group with eight people?

You can, and the transcript will need work. Speaker separation handles a small group reasonably and degrades with numbers, overlap and room reverberation. Recording each participant saying their name at the start helps considerably, and so does a quieter room. Eight people in a seminar room with a hard floor is hard audio for any system.

Is a local transcript GDPR compliant?

Compliance is a property of your processing, not of a tool. What the app removes is the transfer and the processor relationship for the transcription step, which are two of the harder parts. Lawful basis, participant information, special category conditions, retention and security of the device all remain your responsibility and your institution's.

Does it work for longitudinal projects over several years?

Cautiously. There is no server, so continuity depends entirely on your own backups and on keeping devices working over the life of the project. For anything multi-year I would export transcripts into your institutional store as you go, rather than treating the phone as the archive. The app is a capture tool, not a repository.

Before the next round of fieldwork, record a fifteen minute pilot interview with a colleague and transcribe it offline. You will learn more about the accuracy you can expect, and the correction time you should budget, than from any comparison table including mine.

Get Just Transcribe

Or first: How Advanced Privacy works · The full fact sheet

Nine other jobs

Written by the developer, and not an ethics opinion. I make Just Transcribe. Your ethics committee, your institution's data protection office and your funder's data management requirements all outrank this page. What it can do is describe accurately what happens to a recording, so you can write a true sentence about it in a protocol.

Written 26 September 2026, updated 3 October 2026 for Just Transcribe 1.36. Numbers and sources on the fact sheet.