Which transcription app should you pick?
This page is written by the person who makes one of them, so read it with that in mind. It is still an honest guide, including a section on where this app is the wrong answer, because a comparison that never says that is not a comparison.
The one question that decides everything else
Transcription apps look alike from the outside. You press record, you get text. The difference that matters is invisible from the screenshots: where does the speech model actually run?
There are only two answers, and every other difference follows from which one an app picked.
| On a server | Your audio is uploaded, a machine somewhere transcribes it, the text comes back. This is how most transcription products work, including well known cloud services such as Otter, Rev, Trint and Notta. It buys real capabilities: live captions, meeting bots, team workspaces, sync. |
|---|---|
| On your device | The model is downloaded to your phone once and does the work there. Nothing is uploaded, because there is nowhere to upload it to. This is how Just Transcribe works. It gives up sync, teams and live captions in exchange. |
Neither is a scandal. They are trade-offs, and which one is right depends entirely on what you record. A podcast intro and a conversation with a source are not the same recording, and they do not deserve the same tool.
On-device and cloud, side by side
An honest split. The right-hand column is not a list of villains; it is a list of things you are paying for when you pay monthly.
| Where audio goes | On-device: nowhere. It stays in the app's storage on your phone. Cloud: to a server, where it is processed and usually stored for some period. |
|---|---|
| Works without internet | On-device: yes, completely, including airplane mode. Cloud: no. No connection, no transcript. |
| Account | On-device: not needed. Cloud: required, because the server has to know whose file it is. |
| Typical price | On-device: one purchase, or free. Cloud: a monthly fee, often with a cap on minutes. |
| Minutes limit | On-device: none. The limit is your battery and storage. Cloud: usually a monthly allowance. |
| Accuracy ceiling | On-device: good to very good, set by the model you choose and the phone you own. Cloud: higher headroom on hard audio, because a server has more compute to spend. |
| Speed on long files | On-device: slower on an hour-long recording, and it warms the phone up. Cloud: faster, because it is not running on a battery. |
| Live captions | On-device: not in this app. Cloud: commonly yes, and it is genuinely useful. |
| Meeting bots | On-device: no. Nothing joins your Zoom call. Cloud: yes, that is a flagship feature. |
| Team workspace | On-device: no. One person, one device. Cloud: yes, with sharing, comments and permissions. |
| Sync between devices | On-device: no, and that is a direct consequence of not uploading. Cloud: yes. |
| Data processing agreement | On-device: no third party is involved in transcription, so there is nothing to appoint for that step. Cloud: you need one, and you need to read it. |
| If the company disappears | On-device: the app keeps working. The model is already on your phone. Cloud: the service stops and your archive goes with it. |
What “offline” actually means, and how to test it
Plenty of apps say private. Fewer mean it in the strong sense, and the word offline is used for at least three different things:
- Offline recording. The app can record with no signal, then uploads the file when you are back online. The transcription is still done by a server.
- Encrypted in transit. A genuinely good thing, and completely different from not sending it. Your audio still arrives on somebody's machine.
- Offline transcription. The model runs on your device and no audio is ever transmitted. This is the one that survives airplane mode.
You do not have to take anyone's word for which kind you are holding, this page included. The test takes a minute:
| 1 | Install the app and download whatever it needs, over Wi-Fi. |
|---|---|
| 2 | Turn on airplane mode and switch Wi-Fi off as well. |
| 3 | Record something and transcribe it. |
| 4 | If you get a full, accurate transcript with no connection, the model is on your device. If it spins, queues or errors, it was never local. |
Just Transcribe passes this test by design. After the one-off model download it never needs a connection again. In the app's own words, the internet is used exactly twice in its lifetime: once to fetch a model you asked for, and once for the App Store purchase. Never for your recordings. The full breakdown is on the fact sheet and in the privacy policy.
What it actually costs, over time
A monthly fee is easy to say yes to and hard to notice afterwards. Laid out over a few years the difference stops being a rounding error.
| Just Transcribe | Free to download. Tiny and Base free forever. EUR 7.99 once for everything else. Year one: EUR 7.99. Year three: EUR 7.99. Year ten: EUR 7.99. |
|---|---|
| A EUR 8 a month plan | EUR 96 in year one. EUR 288 over three years. |
| A EUR 15 a month plan | EUR 180 in year one. EUR 540 over three years. |
| A EUR 20 a month plan | EUR 240 in year one. EUR 720 over three years. |
The subscription rows are illustrative price bands to show the shape of the maths, not a claim about what any particular product charges. Check the current price of anything you are comparing; they move.
What the monthly fee buys, when it is worth it, is real: servers that keep running, a team plan, live captions, integrations, and someone to call. If you use those, a subscription is not a rip-off, it is the correct purchase. If you were only ever going to record interviews and read them back, you are renting a server you never asked for.
Who is in charge of the model
This is the part most comparisons skip, and it is the difference between using an app and being handed one. With a cloud service the model is whatever the provider decided this quarter. It can change under you, get more expensive, get worse for your accent, or get discontinued.
Just Transcribe hands that switch to you. Ten speech models, and you pick which of them exist on your phone:
- You choose what to download. Nine OpenAI Whisper models, from Tiny at 75 MB to Large v3 Turbo at 954 MB, plus NVIDIA Parakeet v3 at 700 MB, which is very fast across 25 European languages.
- You choose the trade-off. Small and instant for a voice memo, large and slow for a two-hour interview that has to be right. Nobody makes that call for you.
- You can delete any of them. One tap, storage back. The app does not silently keep a copy.
- Nothing changes without you. The model on your phone today is the model on your phone next year, unless you swap it.
- Two are free forever. Tiny and Base need no purchase at all, so the free tier is a working app rather than a countdown.
The complete model table, with sizes, speeds and what each one is good for, is on the fact sheet.
The GDPR, plainly
No app is GDPR compliant on its own, and anything claiming otherwise is selling you something. Compliance is a property of your process, not of a download. What local processing does do is remove a clearly bounded part of the problem.
| Gone | There is no processor for the transcription, so for that step there is no data processing agreement to sign and no transfer outside the EU to assess. |
|---|---|
| Gone | No account and no server storage, so no breach at a supplier that contains your recordings. |
| Gone | Nothing is used to train models. Nothing leaves, so there is nothing to train on. |
| Remains | Are you allowed to record this conversation, and do the people in it know? That does not change with where the software runs. |
| Remains | How long do you keep the recording, and who can get into your device? Your phone is now the archive. |
| Remains | The rules of your own organisation, profession or ethics committee. |
This is a description of how the app works technically, not legal advice. For a specific situation, ask your data protection officer or your own lawyer.
Who ends up choosing on-device, and why
The pattern is consistent: people whose recordings would be a problem in someone else's hands, and people who are tired of renting.
| Journalists | Source protection is not a preference. An interview that never leaves the phone cannot be subpoenaed from a service, leaked from a breach, or read by a support engineer. |
|---|---|
| Healthcare and therapy | Session notes and patient conversations, where uploading the audio would mean a processor, an agreement and a risk assessment for every voice memo. |
| Legal and HR | Privileged conversations, grievance interviews, witness statements. Confidentiality that cannot be delegated to a third party's uptime page. |
| Researchers | Interview studies with ethics approval that specifies where the data may live. Local processing is far easier to get past a review board. |
| Public bodies | Council and board meetings, where procurement and data location questions arrive before the app does. |
| Students | Lectures, all semester, on a budget that does not stretch to a subscription. Unlimited minutes matter here more than anywhere. |
| Field work | Basements, aircraft, trains, remote sites, other countries. No signal is a normal Tuesday, and a cloud app is a brick. |
| People who just object | No confidentiality duty at all, simply not willing to hand a private conversation to a company, or to pay every month for the privilege. |
Where this app is the wrong answer
If any of these describe you, a cloud service is the better purchase and you should make it. None of this is a roadmap being coy; these are consequences of doing the work on one device with no server.
- You need live captions while somebody is still talking. Transcription here happens after the recording, not word by word during it.
- You need a meeting bot. Nothing joins your Zoom or Teams call and takes notes for you.
- You work as a team. There is no shared workspace, no comments, no permissions. One person, one device.
- You want transcripts on every device. Nothing syncs, because nothing is uploaded. Your iPad does not see your iPhone's recordings.
- You are not on Apple hardware. No Android, no Windows, no web version. iPhone, iPad and Apple Watch only.
- You process hours of difficult audio daily. A phone will do it, but a server will do it faster and with more headroom on bad recordings.
- Your phone is nearly full. The good models are large. Large v3 Turbo alone is 954 MB, and it has to live somewhere.
- You want a recovery dashboard. There is no server, so there is nothing to restore from if you delete a recording and have no backup.
And one more, in fairness: if Apple's built-in dictation and Voice Memos transcription already cover what you do, use them. They are free, already installed, and also run on the device. A dedicated app earns its place when you need the model choice, the 50 languages, speaker identification, editing, or export to SRT, VTT and PDF.
Questions to ask any transcription app before you trust it
Useful whether or not you end up here. Every one of these has a checkable answer, and an app that will not answer plainly has told you something.
| 1 | Does it transcribe in airplane mode? The single fastest way to find out where the model lives. |
|---|---|
| 2 | Is there an account? If there is, something is on a server. That is not automatically bad, but it is the answer to the previous question. |
| 3 | What is the App Store privacy label? “Data Not Collected” is a public, auditable claim made to Apple, not a marketing line. |
| 4 | How long is audio kept, and where? A cloud service should be able to tell you in one sentence, with a region. |
| 5 | Is my audio used to train anything? Ask directly. Look for the opt-out and check whether it is on by default. |
| 6 | Which model is it, and can I choose? If the answer is vague, you cannot reason about accuracy and it can change without notice. |
| 7 | What happens when I stop paying? Do the transcripts stay readable, export cleanly, or vanish with the plan? |
| 8 | Can I get my text out? Plain text, SRT, VTT, PDF. Export is the difference between owning your work and visiting it. |
| 9 | Is there a minutes cap? Per month, per file, per upload. Find it before the deadline finds it for you. |
| 10 | Who is behind it? A named company, a real support address, a working privacy policy. Not a form. |
For the record, Just Transcribe's answers: yes it works in airplane mode; there is no account; the label is Data Not Collected; audio is kept only on your device for as long as you keep it; nothing is used to train anything; you choose from ten named models and can delete them; there is nothing to stop paying for; export is SRT, VTT, PDF and text; there is no minutes cap; and it is made by The Dutch Logic Co. in the Netherlands, reachable at support@thedutchlogic.nl.
Straight answers
What is the best offline transcription app for iPhone?
There is no single best app for everyone, so the honest answer is: the best one is the one whose speech model actually runs on your phone, and there are very few of those. Just Transcribe is one of them. It runs nine Whisper models and Parakeet v3 locally, works in airplane mode, needs no account, and costs EUR 7.99 once.
If your test for “best” is accuracy with unlimited compute, a cloud service still has an edge on very long or very noisy files. If your test is that the recording never leaves your hands, on-device is the only category that passes, and price and control decide between the apps inside it.
Is there an app that keeps nothing at all afterwards?
Yes, and the difference is worth knowing. On-device transcription means nothing is uploaded. It does not mean nothing is stored: your recordings still sit on your phone until you delete them, which is exactly what you want for work you intend to keep.
For the stronger case, Just Transcribe has Advanced Privacy: a mode that runs one recording through transcription and a summary and then wipes all of it. The recording never enters the library, the audio is erased the moment the transcript exists, and the rest goes when you close the screen. Export what you need before you leave, because nothing can be recovered afterwards.
Is there a transcription app that does not use the cloud at all?
Yes. Just Transcribe does all speech recognition on the device. There is no server behind it and no upload step. It connects to the internet exactly twice: once to download a model you picked, and once for the App Store purchase. After that, airplane mode changes nothing.
Do not take anyone's word for it, including this page. Run the airplane mode test on any app that claims this.
Is on-device transcription as accurate as a cloud service?
It depends which model you pick, which is exactly why you get to pick. Large v3 Turbo, at 954 MB, is OpenAI's Whisper, the same model cloud services run, and on a recent iPhone it lands in the same league on ordinary speech. The small models trade accuracy for speed, and Tiny is noticeably worse.
Where a server genuinely wins is headroom: very long recordings, very poor audio, heavy overlapping speech. A phone runs on a battery and a server does not.
What is a good alternative to a transcription subscription?
A one-time purchase, if your work fits inside it. EUR 7.99 once, against EUR 360 over three years at ten euro a month. What you give up is the team workspace, the meeting bot, live captions and sync between devices. If you use those, keep paying, because that is what the fee is for. If you never have, you are renting a server you never asked for.
Can I transcribe confidential recordings without sending them to a third party?
With an on-device app, yes. Here the audio, the transcript and the speaker voiceprints stay in the app's private storage on your phone, and no third party takes part in the transcription at any point.
That is a description of how the software works, not legal advice. Your own consent, retention and duty-of-care obligations do not disappear because the processing happens locally.
Which transcription app is GDPR compliant?
None of them, on its own. Compliance is a property of your process, not of a download, and anything claiming otherwise is selling you something.
What an on-device app removes is one specific piece of the problem: nothing is uploaded, so for that step there is no processor to appoint, no data processing agreement to sign and no transfer to assess. Just Transcribe collects nothing and carries the App Store label Data Not Collected. See the GDPR, plainly for the full split.
Does the transcription built into the iPhone do the same thing?
Partly, and if it covers what you do, use it. Apple's dictation and Voice Memos transcription are free, already installed, and also run on the device.
A dedicated app differs on control and output: ten models with a known accuracy and size trade-off, 50 transcription languages, speaker identification with a saved voice library, transcript editing with find and replace, and export to SRT, VTT and PDF.
Can I run OpenAI Whisper on my iPhone?
Yes. Whisper is an open model and it runs on modern iPhones through Core ML. This app ships nine Whisper models via WhisperKit, plus NVIDIA Parakeet v3 via FluidAudio. You choose what to download and can delete any of it again.
If you are technical and own a Mac, you can also run Whisper yourself for free with whisper.cpp. That is a genuinely good option when the recording is already on a computer, and a poor one when it is a conversation you are having in ten minutes.
When is Just Transcribe the wrong choice?
Live captions, meeting bots, team workspaces, sync between devices, or any hardware not made by Apple. The full list is in where this app is the wrong answer, written without softening.
Next
If the on-device column is the one you want, the download is free and two models stay free forever, so you can run the airplane mode test yourself before spending anything.
Get the app Read the fact sheet
Verified 27 August 2026 against app version 1.33. Prices in euro, including VAT where applicable, as shown on the European App Store.