Just Transcribe.
Advanced Privacy
Just Transcribe: Offline is a voice-to-text app for iPhone, iPad, Mac and Apple Watch, with an Android app on Google Play. It records and transcribes with speech models that run entirely on your device (21 of them, from OpenAI's Whisper to Meta's Omnilingual), so your recordings never leave it. Works fully offline. No cloud. No account. No tracking.
DaanWe shipped the onboarding fix this morning.
SaraThen the release moves to Friday, so QA gets the full day.
DaanI'll send the updated timeline right after this call.
Recording for work? Lawyers, doctors, therapists and seven more →
Group recordings into folders, and folders into folders, five levels deep, each in a colour its subfolders follow. Then search by title or by the words inside any transcript, straight through every folder below.
Fifty-nine languages plus automatic detection, now with Japanese, Korean, Cantonese, Thai, Vietnamese, Indonesian, Malay, Filipino and Persian. Meta's Omnilingual model goes further, to more than 1,600 languages without asking which. A finished transcript translates into 55 languages on the device, and the app itself speaks 14 languages.
Twenty-one models is more than anyone can pick from blind. Answer four questions (which language, how many voices, speed or accuracy, follow along or not) and the app shows the one that fits, with a quicker and a sharper alternative and the best free option.
Omnilingual writes in lower case without a single comma. One tap and the on-device language model puts the capitals and punctuation back, without changing a word: if it tries to, that passage stays exactly as it was.
Start a recording from your car's screen with Apple CarPlay, hands on the wheel and eyes on the road. It's waiting on your iPhone when you stop.
Recorded from the back of the room? One tap cleans up the audio first: soft voices get lifted, low rumble gets cut, and the transcript comes out sharper.
The Mac app doubles as the fast lane. Pair it once with your iPhone over your own Wi-Fi, send a long recording across, and the Mac works through it far quicker with the same 21 models, then sends the text back. It records and transcribes on its own too, on macOS 26 or later. Two machines you own, with no transcription service in between.
Switch it on in Settings and ask Siri for your latest recording, a transcript, a summary or the action items, or have it search everything you recorded. Recordings show up in Spotlight too. It all stays in your iPhone's own index, and it stays off until you turn it on.
Pick a speech model, or answer four questions and let the app pick one, and it downloads to your device, once. This is the bit that turns speech into text, and it runs fully on your phone. Smaller is faster, larger is more accurate.
Pick the model and language right on the main screen. Choose the exact language for best accuracy, or leave it on Automatic. Everything else lives in one tidy Settings menu.
Tap Start recording to capture with the microphone, or import an existing audio file (wav, mp3, m4a, aiff and more).
Open a recording and tap Transcribe. The model converts the audio to text right there on your device. The Options menu can also enhance the audio first, label the speakers, or run it again with a different model.
Read the full transcript, copy it to the clipboard, or share it straight into any other app.
Twenty-one speech models from five engines: nine from the Whisper family (six released by OpenAI, three distilled or optimised by others), NVIDIA's Parakeet v3 and Canary 1B v2, Qwen3-ASR for the Asian languages and Meta's Omnilingual ASR for more than 1,600 languages. Downloaded once and kept on your device. Each one shows its quality at a glance, from Basic to Best, and four questions point at the one that fits.
Blink and it's done. Just don't be shocked when "good morning" comes out as "banana".
The sensible one. Reliable and unremarkable, the Toyota Corolla of transcription.
The finest there is, turbo and all. Your phone whispers a quiet prayer, then absolutely delivers.
Parakeet's studious cousin. The same 25 European languages, rather better at punctuation, names and the messy bits, and it takes its sweet time.
The one that actually speaks Asia. Chinese, Cantonese, Japanese, Korean, Thai, Vietnamese and a dozen more, properly. Takes its time about it.
The featherweight of a very unusual family. One model for over 1,600 languages, and it never asks which one.
…plus Small, Large v2, Large v2 Turbo and Large v3, the English-only Distil models, NVIDIA's very fast Parakeet v3, the sharper Qwen3 ASR 1.7B and seven more Omnilingual builds, up to 7 GB. You decide which ones land on your phone, keep several side by side, and delete any of them again. All 21, with sizes and speeds →
The same app on the iPad and on the iPhone, in daylight and in the dark. Nothing has to be important to be worth recording, and nothing has to go anywhere to be typed out.
Most transcription apps ship your audio off to a server. Just Transcribe never does: the AI runs on your phone, so there's nothing to upload in the first place, and the whole thing works fully offline.
"I wish the world cared a little more about privacy and quality. I'm not a big company. I simply needed a tool that turns my voice into text entirely on my own device, with nothing sent to the cloud. So I built it.
That other people get to enjoy it too? That's just a bonus."
Daan · The Dutch Logic Co.
The app already does all of its thinking on your phone. Advanced Privacy is the mode where you can watch that happen: one recording that never reaches your library, the audio erased the moment the words have been read out of it, and everything that is left wiped when you close the screen.
Ten jobs where the person on the recording never chose the software. Each one has a page of its own: what you would actually use it for, and where I would tell you to buy something else.
Yes, completely. The AI model lives on your phone, so transcription works in a tunnel, on a plane or with your data switched off. The only time the app goes online is to download a model, once.
No. Audio and transcripts are processed and stored on your iPhone, iPad, Mac or Apple Watch. There is no server to upload to, no account to create and no analytics.
Yes. Advanced Privacy is a separate mode that runs one recording from start to finish and then wipes it: nothing reaches your library, the audio is erased the moment the transcript exists, and the rest goes when you close the screen. It works in airplane mode, and it is part of the one-time Pro unlock.
The download is free, and the Tiny and Base models stay free forever, so you can record, import and transcribe without paying anything. A new install also gets a free hour on every other model, to hear the difference first. A single one-time unlock of EUR 9.99 then opens every other feature, permanently: no subscription, no ads, nothing monthly.
Fifty-nine languages plus automatic detection, from Dutch, English, German, French, Spanish and the Nordic languages to Arabic, Hindi, Chinese and Russian, and since version 1.36 also Japanese, Korean, Cantonese, Thai, Vietnamese, Indonesian, Malay, Filipino and Persian. Pinning the exact language gives the sharpest results. Meta's Omnilingual model goes further still: more than 1,600 languages, without being told which. A finished transcript can also be translated into 55 languages, still on your device, and the app's own interface is available in 14 languages.
Yes. Speaker identification runs on your device and labels who said what in the transcript. Save a voice once and returning speakers are named automatically.
Yes. There is an Apple Watch app and a watch face complication, so you can start a recording from your wrist and it is waiting on your iPhone afterwards. You can start one from CarPlay too.
wav, mp3, m4a, aiff and more. Import a file you already have and transcribe it the same way as a fresh recording.
That depends on the model you pick. There are 21, from Tiny (75 MB, fast and rough) to six that the app rates 5 out of 5, among them Large v3 Turbo (954 MB), which is OpenAI's own Whisper large-v3-turbo rather than a cut-down phone build, and NVIDIA's careful Canary 1B v2. Four questions in the app point at the one that suits your language and your recording.
No. There is no sign-up, no login, and the app collects no personal data at all.
Free to download, and free to keep using with two real models. One in-app purchase of EUR 9.99 unlocks everything else, once, forever. No subscription, no ads, no account. The Mac app is free to start too, and has its own one-time unlock of EUR 19.99.
The recording and the transcript never leave your device, and there is no server that could receive them, so there is no transfer to a processor to cover. Whether you may record a conversation, and who has to be told, is still up to you and the same rules as before.
Twenty-one, from five engines, from 75 MB to 7 GB. You pick which ones land on your phone, keep several side by side, switch per recording, and delete any of them again. Nothing downloads without you asking. See the full list
Yes. A transcript exports as SRT or VTT subtitles, as a PDF, as a text file or Markdown with speaker labels, or as lines with timestamps.
Let the app answer that: the model manager asks four questions (which language, what you are recording, speed or accuracy, follow along or just text) and 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, Qwen3-ASR for Chinese, Japanese, Korean and the rest of East and Southeast Asia, and Omnilingual for a language nothing else covers.
On Android and Mac, yes. The Android app is on Google Play. On Apple devices: iPhone on iOS 17 or later, iPad on iPadOS 17 or later, Mac on macOS 26 or later, Apple Watch on watchOS 10.6 or later, plus CarPlay. The Mac app is free to start, like the one on the iPhone, and has its own one-time unlock of EUR 19.99: buying Pro on the iPhone does not unlock the Mac, or the other way round. There is no web version.
Just Transcribe is free on the App Store, and stays free with two real speech models, so this is not a trial that runs out. A new install also gets a free hour on every other model. A single one-time unlock of EUR 9.99 then opens every other feature: no subscription, no ads, nothing monthly. Pay once, keep it for good. The Android app is on Google Play, and the Mac app is in the Mac App Store: free to start as well, with its own one-time unlock of EUR 19.99.
