How can we help?

First time only: download a model (see the next question). Then tap Start recording and speak. The blue button becomes a coral dot and a live waveform shows it’s listening. Recording keeps going even if you lock the screen or switch to another app, so a long session isn’t cut off; iOS shows a recording indicator until you tap stop.

Your recording appears in the list. Tap Transcribe on it and the model turns the audio into text, right on your device.

The “model” is the AI that converts speech to text. Just Transcribe has 21 of them, from five engines (the Whisper family, NVIDIA’s Parakeet and Canary, Qwen3-ASR and Meta’s Omnilingual), all running entirely on your device. The model files have to live on your phone for that to work offline, so you download one the first time (tap the download icon in the top-right).

Smaller models (from ± 75 MB) are quick; larger ones are more accurate but bigger and slower. You can keep several installed and switch between them anytime. Not sure which? Find your model, at the top of the model manager, asks four questions and points at the one that fits.

Open the model manager and tap Find your model: four questions (which language, what you are recording, speed or accuracy, follow along or just text) and it shows the model that fits, with a quicker and a sharper alternative and the best free one.

As a rule of thumb: Large v3 Turbo for most languages, Parakeet v3 when speed matters in a European language, Canary 1B v2 for the most careful European text, Qwen3-ASR for Chinese, Japanese, Korean, Thai, Vietnamese and their neighbours, and Omnilingual for a language nothing else covers. Canary and Qwen3-ASR give text only, so no follow along or speaker labels. All 21 are on the fact sheet.

That is Omnilingual: it was trained to write in lower case without punctuation, and that is the price of knowing more than 1,600 languages. Open the Options menu on the recording and choose Add punctuation. The on-device language model puts the capitals and punctuation back without changing a single word; a passage where it tries to change one is left exactly as it was. It needs one of the summary models on the device, and it is part of Pro.

Yes, once you switch on Let Siri use my recordings in Settings; it is off until you do. Then you can ask Siri to search your recordings, read a transcript, give a summary or the action items, or open the latest one, and recordings show up in Spotlight on iOS 18 and later. Everything goes into your iPhone’s own index and stays there, and switching it off removes it again. Siri cannot start a recording.

Once you’ve downloaded a model, yes. Recording and transcription run fully offline, in airplane mode or anywhere with no signal. The internet is only needed the one time you download a model (and when you make the one-time unlock purchase).

No. Transcription runs on the device, so your audio and transcripts never leave your phone. There are no servers and no account. The only thing the app ever downloads is the model itself, never your content. Details are in our privacy policy.

Fifty-nine languages, plus Automatic if you want the model to detect it (Canary and Qwen3-ASR want the language picked by hand). Omnilingual is different again: it covers more than 1,600 languages and never asks which one. Picking the exact language gives the sharpest results, and you can change it at any time and re-run a transcript with Again if you picked the wrong one. A finished transcript can also be translated into 55 languages, on the device. The app’s own interface is available in 14 languages. The fact sheet lists them all.

Use Advanced Privacy, the lock in the top left corner of the main screen. It runs one recording all the way through and then wipes it: nothing is written to your library, the audio is erased the moment the transcript has been read out of it, and the transcript and summary go when you close the screen.

Before you start you need Large v3 Turbo and a summary model on the device, and enough battery (stay above 40% or plug in). Downloading those models is the only step that needs a connection, so you can switch on Airplane Mode afterwards and watch the whole session run offline. Export the summary before you leave: once you go, nothing can be recovered. Full explanation on the Advanced Privacy page.

A few things help:

• Use a larger model (Small and up are noticeably better than Tiny), or let Find your model pick one for your language.
• Set the exact language instead of Automatic when you know it.
• Record in a quiet spot, close to the speaker, and ask people not to talk over each other.
• For a quiet or noisy recording, run Enhance audio from the Options menu first.

It cleans up the recording before transcribing: soft voices get lifted and low rumble gets cut, which usually gives a noticeably sharper transcript. Run it from the Options menu on a recording. Changed your mind? Undo enhance puts the original back.

Yes. Export gives you SRT subtitles, VTT subtitles, a PDF, a text file or Markdown with speaker labels, or lines with timestamps to paste anywhere. It sits in the Options menu on any recording that already has a transcript.

Yes, into 55 languages, and still entirely on your device. Translation uses a separate local model that you download once, the same way as a speech model. Whisper can also translate speech straight into English while transcribing, on the multilingual Whisper models (not on Distil, Parakeet, Canary, Qwen3-ASR or Omnilingual).

Yes. Tap the import button next to the record button and pick an audio file: wav, mp3, m4a, aiff and more are supported. The app transcribes that single file on your device; it has no access to anything you don’t choose.

The very first time you use a model on a device, iOS optimises it for your specific hardware, which can take a little while. After that it loads quickly. Bigger models also simply take longer per minute of audio, and the app shows an estimate before you start.

Group recordings into folders to keep things tidy, and put folders inside folders, up to five levels deep, each in a colour its subfolders follow. The search box finds any recording by its title or by the words inside its transcript, and a search inside a folder also looks in the folders within it.

Delete any recording or transcript from inside the app and it’s gone from your device. To reclaim storage, remove models you no longer use from the model manager. Deleting the app clears all of its local data. (If you use iCloud or device backups, those may keep a copy until the backup updates.)

This is almost always a permission issue. Go to Settings → Just Transcribe and make sure Microphone is turned on. (It is the only permission recording needs.) If it was denied earlier, switching it on here fixes it; restarting the app afterwards helps too.

Just Transcribe is free to download, and the Tiny and Base speech models stay free forever, so recording and transcribing cost nothing at all. A new install also gets one free hour on every other model, to try the big ones first. A single one-time unlock of EUR 9.99 then opens every other feature, permanently. It’s a one-time purchase, not a subscription: no monthly fees, no ads and no extra charges after that. The Mac app works the same way, with its own one-time unlock of EUR 19.99: buying Pro on the iPhone does not unlock the Mac, or the other way round. The App Store shows the exact price for your region. Refund requests go through Apple at reportaproblem.apple.com.

Yes. When you play a recording, the transcript highlights line by line so you can read along, and you can tap a line to jump straight to that moment. You can also edit the text directly to fix a name or a number, and use find & replace to correct the same word everywhere in one go.

Yes. You can start a recording straight from your Apple Watch, or from its complication on the watch face, without reaching for your phone. The recording is then waiting on your iPhone, where you transcribe it on the bigger screen.

iPhone with iOS 17.0 or later, iPad with iPadOS 17.0 or later, Mac with macOS 26 or later, and Apple Watch with watchOS 10.6 or later. There is also an Android app on Google Play. Larger models need more free storage and run faster on newer devices. Canary 1B v2 needs iOS 18, and the largest Omnilingual builds only show up on a device with enough memory to run them.