For human rights work. The testimony should outlive the recording.
Interviews with witnesses and survivors, taken in places with no network, carried across borders where devices are searched. The words need to survive. The audio, usually, should not.
Use it forWitness and survivor interviews, work with no network, and anything you would rather not carry across a border as audio.
The argumentAdvanced Privacy erases the audio the moment it becomes text, so the recording of a voice exists for minutes instead of years.
Not forA device that is taken and unlocked. The transcripts are on it, and that is the risk this app helps with least.
Each of these assumes you have taken informed consent properly, in a form the person actually understood, and that they know what will be kept.
01Witness and survivor interviews
Long, difficult, and often in a language you will need help with later. Taking notes by hand while someone describes the worst day of their life is both slow and a barrier between you and them.
A recording that becomes text within the hour, and then stops being a recording, is a better shape for this work than a file that travels home.
In the app: record with Advanced Privacy so the audio goes as soon as the transcript exists. That mode has no editor: if names have to come out of the text before it leaves the device, record in the ordinary mode instead, remove them with find and replace, and delete the recording right after transcribing.
02Field work with no network at all
Camps, remote districts, places where the network exists and should not be trusted. Cloud transcription is not slow here, it is absent, and the version that queues an upload for later is the most dangerous kind.
In the app: keep the device in airplane mode for the entire deployment. Recording, transcription, translation and summarising all work without any radio at any point.
03Interviews in a language you do not read
Transcribe in the original language first, so the record of what was said exists in the words that were used. Machine translation then tells you which testimony needs a trusted human translator, which is a decision about whose time and trust to spend.
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. The original stays the record, the translation is triage.
04Crossing a border with the material
The moment where everything that exists on the device can be examined. What you carry should be the minimum that is useful, in the least identifying form.
Audio erased at the point of transcription means the biometric version of the testimony is already gone before anyone asks.
In the app: Advanced Privacy plus a deliberate export and delete routine before travel. And a real security plan, because this is the risk the app helps least with.
05Case notes and legal aid intake
Asylum interviews, legal aid intake, documentation for a claim. Long factual accounts where dates and sequence matter, and where the person telling it is exhausted and will not want to repeat it.
In the app: transcribe locally, correct in the editor, and search the full text later for the date or place that becomes important once the file is built.
The recording is often more dangerous than the testimony
A written account can be anonymised, redacted and carried across a border in a form that identifies nobody. An audio recording cannot: a voice is a biometric identifier, and a recording of a person describing what happened to them is proof that they spoke to you, which in some contexts is the dangerous fact entirely independent of what they said.
So the goal is to hold the audio for the shortest time that is useful and then not have it, which is a stronger requirement than simply not uploading.
What the ordinary workflow creates
An audio file that travels home in a bag, sits on a laptop, gets uploaded to a transcription service for speed, and ends up in a cloud folder shared across an organisation. Every one of those steps is a place it can be found, and the border crossing is the one where finding it has consequences for the person who spoke.
Transcription services also usually sit in jurisdictions with their own disclosure regimes, and they cannot promise you what they would do under compulsion.
What Just Transcribe changes
The transcript is made on the device, in the field, without a network. With Advanced Privacy the audio is erased as soon as the transcript exists and the session is cleared when you leave the screen, so the window in which the recording exists is minutes rather than months.
There is no account, no upload and no analytics. The app opens a connection twice in its life: to fetch a model you chose, and for the purchase.
This shortens the life of the most dangerous artefact. It does nothing about the transcript, your notes, your travel, your contacts, or a device that is taken from you and unlocked. Those are the larger risks and they need a plan that is not an app.
Why Just Transcribe
Two of these are specific to this kind of work and are the reason the page exists.
✓The audio can be erased at the moment it becomes text. Advanced Privacy records once, deletes the audio as soon as the transcript exists, and wipes the session on leaving the screen.
✓No network call, ever, after setup. Nothing uploads, nothing queues, nothing retries when a signal appears at the wrong moment.
✓No account. No identity, no login, no record anywhere that ties you to an interview.
✓Nobody to compel. A company that never received the file cannot be ordered to produce it or gagged about it.
✓Works anywhere, in any of fifty-nine languages, and more than 1,600 through Omnilingual. With automatic detection, which matters when you do not know in advance what language someone will feel safest speaking.
✓Editing on the device. Remove names with find and replace before the transcript is exported anywhere.
✓One payment, no renewal. Nothing expires mid-deployment and nothing calls home to check a licence.
Advanced Privacy turns the recording into a transcript and then deletes the recording, so the biometric version of a person's testimony exists for minutes rather than for years.
In this work that is a materially different risk profile, and Advanced Privacy is the part of the app built for it.
What this helps with, and what it does not
Where this helps, where it does not, and what still needs a plan.
A rough model, not advice. If consequences are severe, work with a digital security adviser.
Interception of an upload
Addressed. There is no upload, so there is nothing to intercept and no traffic revealing that an interview took place.
Vendor compulsion or breach
Addressed. No company receives the audio or the transcript, so there is nobody to serve and nothing to leak.
Audio recovered later
Largely addressed with Advanced Privacy: the audio is erased at transcription. Deleted files on modern encrypted devices are hard to recover, though no consumer app can promise forensic erasure.
Device search or seizure
Not addressed. Transcripts are on the device. Carry less, export and delete before travel, and get real advice about crossings.
Coercion of you
Not addressed, and no software can be. Fewer artefacts means less to hand over, not immunity.
Informed consent
Not addressed by any tool. The person must understand what is recorded, what is kept, for how long, and what happens if they change their mind.
No security audit of this app has been published. The verifiable claim is behavioural: it completes a full cycle in airplane mode, which means the processing happened on the device.
What the person has to understand before they agree
I am recording this on my phone. It goes to no company and there is no account attached to it. The phone turns it into text and then erases the recording of your voice, usually within the hour. What I keep is the written account, and I can take your name out of it. You can stop now, and you can tell me later to delete it.
In a language they actually speak, and slowly. If you cannot explain it, you do not have consent, and a recording taken without understanding is a risk you created for somebody else.
Before a deployment
Four steps, and the first one is the one that protects people.
Settle consent and the do no harm question first. What is recorded, what is kept, how long, who sees it, and what happens if they withdraw. If you cannot explain that in the language they speak, do not record.
Download the model at home, then stay offline. Pick the largest one the device will hold and test it in the languages you expect. After that, airplane mode for the whole deployment.
Make Advanced Privacy the default. Audio erased at transcription, session cleared on leaving the screen. Use ordinary recording only when you have a specific reason to keep a tape and have thought about what that means for the person on it.
Have an export and deletion routine before travel. Names removed with find and replace, transcripts exported to wherever your organisation actually keeps them, local copies deleted. Do it the same way every time so it does not depend on remembering.
Everything else about device security is outside this app and more important than it: a long passcode rather than a short one, biometrics considered carefully at borders, encryption, and an organisational plan for a device that does not come back. If your organisation has a digital security team, this is their decision, not yours or mine.
What Just Transcribe does not protect against
Read this section twice. In this field the gap between what a tool does and what people assume it does is where harm happens.
×A device that is taken and unlocked. The transcripts are on it. Consumer file protection is not a container with a passphrase only you know, and border procedures are not a technical problem.
×No security audit. No independent review has been published. Verify the behaviour yourself rather than trusting the claim, and treat everything on this page accordingly.
×Not forensic deletion. Advanced Privacy erases the audio and modern encrypted storage makes recovery hard, but no consumer app can guarantee that a determined forensic examination finds nothing.
×No anonymisation. Names are transcribed as spoken. Removing them is manual work you have to remember to do, every time.
×No secure transfer. Getting a transcript to your organisation is entirely your problem, through whatever secure channel you already trust.
×No organisational control. Nothing can be centrally configured, audited or remotely wiped by your security team.
The single useful sentence: it shortens the life of the most identifying artefact and removes the third party. Everything else in this threat model still needs the plan your organisation should already have.
Straight answers
Is this safe to use for testimony in a hostile environment?
It is safer than uploading and it is not a security plan. It removes two risks completely, the vendor and the network, and shortens the life of the audio to minutes. It does nothing about seizure, coercion, your contacts or your movements. If those are live risks for the people you interview, work with a digital security adviser and treat this as one component.
Does Advanced Privacy really delete the audio?
It erases the recording as soon as the transcript exists, stores it in a temporary location with iOS file protection while it exists, and clears the session when you leave the screen. On an encrypted device that makes ordinary recovery very unlikely. It is not a forensic wipe and no consumer app can honestly promise one, so do not carry a device across a hostile border on the strength of that alone. The full description is here.
Can your company be forced to hand over interviews?
No, because we never receive them. There is no server that touches your content, no account and no log of what you recorded. A demand served on The Dutch Logic Co. would find nothing. That is a description of an architecture rather than a promise about behaviour, which is why it is worth more.
Which languages does it handle in the field?
Fifty-nine to choose from for transcription, with automatic detection, and Meta's Omnilingual model recognises more than 1,600 without being told which, many of them languages nothing else on a phone covers. A finished transcript translates into fifty-five on the device. Coverage is uneven: major languages are good, smaller ones and strong regional accents are noticeably worse, and Omnilingual writes without punctuation until you add it back. Test the specific languages you expect before a deployment rather than discovering the gap in a camp with no network.
Should the person being interviewed be told about all this?
Yes, and in terms they can actually use. That a recording is being made, that it will become text and then be deleted, that nothing goes to a company, what will be kept and for how long, and that they can stop. Being able to say truthfully that the audio will be gone within the hour is worth more in that conversation than any feature on this page.
Next
Before a deployment, test the whole cycle at home in the languages you expect: airplane mode, record, transcribe, translate, and confirm that Advanced Privacy leaves you with text and no audio. Knowing exactly what the tool does, and does not do, is the part you owe to the people who agree to talk to you.
Written by the developer, and not security advice. I make Just Transcribe. If the people you interview face serious consequences for having spoken to you, your operational security needs an adviser and a plan, not a recommendation from a software vendor. If your organisation has a digital security policy, that policy wins.
Written 26 September 2026, updated 3 October 2026 for Just Transcribe 1.36. Numbers and sources on the fact sheet.
