Something wrong?

Write to stray@grunt.be. The address carries another Grunt app's name, and it is the right one for Plain Said too. That is the whole of it: an email address, and nothing to sign up for. Bugs, questions, a transcript that came out strangely, or something the app should do and does not: the same address takes all of them, and it is read.

What to put in the message

The answer is faster when the message says which iPhone and which iOS version, and then, for whatever went wrong:

  • which engine — Apple or Whisper — and, for Whisper, which model;
  • which language you had chosen;
  • where the audio came from: recorded in the app, imported from Files, or sent in from another app's share sheet — and roughly how long it is;
  • what you did immediately before it went wrong, and whether doing it again does the same thing.

Please do not send the audio or the transcript unless we ask. They are the private part, we have no need of them to start looking, and the whole point of the app is that they stay with you. If a particular file is the only way to reproduce something, we will say so and you can decide then.

What happens to the message itself is written down rather than left to trust: When you write to us in the privacy policy says what arrives, why we have it, how long we keep it, and who else sees it on the way.

Getting it

On the App Store: Plain Said. Or join the beta on TestFlight. It needs iOS 26.0 or later on an iPhone; it also runs on an iPad, as an iPhone app does, in portrait. It is free, in every country and region except Russia. The app is a few megabytes; a Whisper model, if you want one, is downloaded separately and is much larger.

Getting audio in

Three ways, and they behave slightly differently afterwards:

  • Record it in the app.
  • Import a file from Files.
  • Send it from another app — Voice Memos, Files, Mail — through the share sheet, which takes one audio file at a time.

Anything iOS recognises as audio can be chosen. That is not a promise that everything works: a file in a codec iOS cannot decode will fail with an error rather than transcribe. If that happens, converting it to something ordinary — m4a, wav — and trying again is the quickest route.

A recording keeps running when the screen locks. That is deliberate, and it is why the app declares background audio.

Which engine, and which language

Apple's engine is the default and needs nothing downloaded from us: the first time you use a language with it, iOS fetches that language's files from Apple. Whether it is available at all depends on the device — where it is not, the app says so on the spot instead of failing quietly.

Whisper is the alternative, run on the phone, with a model you download in Settings.

Four languages: English, French, Italian and Dutch. Dutch works on Whisper only — Apple's transcriber has no Dutch on iOS 26 — so a Dutch recording needs a Whisper model downloaded first. The app starts on whichever of the four comes first in your iPhone's language preferences, and English if none of them do.

You choose the language; the app does not detect it. Picking the wrong one gives you a bad transcript rather than a correction or a warning, so it is the first thing to check when a result is nonsense.

Whisper models

Nothing is bundled with the app. In Settings you download the one you want, and the sizes as Hugging Face lists them today are:

  • Tiny — 76.6 MB. Quick and rough.
  • Base — 146.7 MB. The one the app starts on.
  • Small — 486.5 MB.
  • Large v3 Turbo — 645.7 MB. The most accurate, and the slowest.

About 2.8 MB of tokenizer files comes down behind whichever you pick, the first time that model loads. Deleting a model in Settings takes the model and its tokenizer with it.

The first transcription with a new model is the slow one. It has to be prepared before it can run, which can take several minutes; the app tells you while it is happening. After that it is on the phone, and transcribing makes no network request at all.

Getting the text out

Copy Text puts the transcript on the clipboard. Export writes it as plain text (.txt), Markdown (.md) or subtitles (.srt) and hands it to the share sheet, so it goes wherever you send it. You can edit the text before you do — the transcript is editable, and the title is renameable.

Why will this transcript not play?

Tapping a line to hear the audio from that point works for recordings you made in the app, because their audio is still there. Imported and shared audio is deleted as soon as its transcript is done, so those transcripts are text alone. That is the design rather than a fault: the app does not hold on to a copy of a file you already have.

Where did my import go?

An import you never transcribed is removed once it has waited at least seven days — the next time you open Plain Said, or a later time if it cannot then, and that takes its title and any partial transcript with it. Nothing is removed while the app is closed; the clock starts the first time you open the app after the import; and if the import is still being worked on, or anything about the check goes wrong, it keeps your file and tries again later.

Recordings you made in the app are not affected by any of that. They stay until you delete them.

Your recordings are not backed up. Export what matters.

Your recordings and transcripts are deliberately left out of iCloud Backup, and there is no sync. That is the privacy promise working as intended, and it has a price worth saying out loud: deleting the app, or moving to a new iPhone, loses them. There is no way for us to restore anything, because we never had it.

So if a transcript matters, export it. Settings may travel with an iCloud backup the way any app's preferences do, but they are only the engine, the language and the chosen model — never your audio or your text.

Privacy, briefly

Nothing of yours reaches us through the app. No account, no analytics code, no crash reporter of ours, no server of ours. Your audio and transcripts leave the iPhone only when you send them, through the share sheet or the clipboard. What does reach us is Apple's doing rather than the app's: public-link tester data, TestFlight crash reports and App Store information. And then there is whatever you send us yourself, by writing — which the privacy policy accounts for too.

The app can make one kind of request, and only when you tap Download on a Whisper model: the model and its tokenizer come from Hugging Face, which sees your IP address the way any web server does and nothing you recorded or wrote. Separately, iOS downloads Apple's speech languages from Apple the first time you use one. On the default engine with its language already installed, the app makes no network request at all. The privacy policy is the exact version, file by file.

Why does the App Store say it collects location and diagnostics?

Because of the Whisper model download, and only that. The store's privacy label names Location (coarse), Usage Data and Diagnostics. None of it is a location the app asks iOS for — Plain Said never requests location permission — and none of it reaches us. It is what Hugging Face records when the app fetches a model you asked for: the address the request comes from, the approximate place that implies, and the device and system details any web request carries. Stay on the Apple engine and the app makes no network request at all. The privacy policy goes through it line by line.

Why does it sometimes say Transcriba?

The app is called Plain Said everywhere you see it — the Home Screen, the share sheet, and the App Store listing. The project it is built from, its bundle identifiers (be.grunt.transcriba) and the App Store record underneath keep an earlier internal name, Transcriba, and changing those now would buy nothing. So if that name turns up somewhere technical, it is the same app.

← Back to Plain Said