A transcription app for iPhone

Speech in. Plain text out.

Plain Said records what is said, or takes in audio you already have, and turns it into text on the iPhone itself. There is no account, nothing to sign in to, and no server of ours anywhere in it.

Free on the App Store. iPhone, iOS 26 or later — and it runs on an iPad the way any iPhone app does, portrait only.

The app's mark: a line that was spoken, settling into lines that are written.

Four steps,one device.

Audio goes in, text comes out, and the part in between happens on the phone in your hand. The numbers below are the order it happens in, not a ranking.

  1. Get the audio in

    Record it in the app, pick a file in Files, or send it over from another app's share sheet — Voice Memos, Files, Mail. The share sheet takes one audio file at a time.

  2. Say which language

    English, French, Italian or Dutch, and which engine should do the work. You choose both: there is no automatic detection, so the language you pick is the language it transcribes in.

  3. It runs on the phone

    Apple's built-in speech recognition, or a Whisper model you downloaded. Once the model or the language files are on the device, transcribing makes no network request at all.

  4. Read it, fix it, take it out

    Edit the text where it is wrong, rename it, then copy it or export plain text, Markdown or subtitles. With a recording you made in the app you can also tap a line to play the audio from that point — imported audio is gone by then, so those transcripts are text alone.

Two engines.Both on-device.

One is already in iOS and needs nothing from you. The other is a model you choose and download, and it is the only thing in this app that touches the network.

Apple Speech · the default

The speech recognition built into iOS 26. Nothing to download from us; iOS fetches that language's files from Apple the first time you use it. Whether it works at all depends on the device, and where it does not, the app says so rather than failing quietly.

Whisper · if you want it

OpenAI's Whisper models, run on the phone by WhisperKit. None of them ship with the app: you pick one in Settings and it comes down from Hugging Face — Tiny at 76.6 MB, Base at 146.7 MB, Small at 486.5 MB, or Large v3 Turbo at 645.7 MB, as Hugging Face lists them today, with about 2.8 MB of tokenizer behind whichever you pick. Base is the one it starts on.

Which to use

Accuracy depends on the audio, the language and the model. Tiny is quick and rough; Large v3 Turbo is the most accurate and the slowest. Dutch is Whisper only, because Apple's transcriber has no Dutch on iOS 26.

The first run is the slow one

A Whisper model has to be prepared before its first transcription, and that can take several minutes. The app tells you it is doing it. After that the model is on the phone and the network is out of the picture.

What it will notdo for you.

Worth knowing before you rely on it, rather than after.

Your recordings have no second copy

Your recordings and transcripts are on this iPhone and nowhere else: they are deliberately kept out of iCloud Backup. So deleting the app, or moving to a new iPhone, loses them unless you exported what you wanted first. There is no sync and no second copy.

Imported audio does not stay

Audio you import is deleted once it has been transcribed; the transcript stays. If an import is never transcribed, Plain Said removes it once it has waited at least seven days — the next time you open the app, or a later time if it cannot then. Audio you recorded in the app is the other case: that is kept until you delete the recording.

It does not know who is speaking

There are no speaker labels. A transcript is one run of text with timestamps, not a conversation split between people.

Four languages, chosen by you

English, French, Italian and Dutch, and no automatic detection — pick the wrong one and you get a wrong transcript rather than a correction. The interface itself is in English only.

Your voice isnot the product.

A recording of someone talking is about as personal as a file gets. Plain Said has no account to attach one to, no analytics code, no tracking and no server of ours to send one to.

Audio and text stay put

Both live in the app's own storage on the iPhone. The only way either leaves is the way you send it: the iOS share sheet, or the clipboard.

One download, nothing else

On the default engine the app makes no network request at all. Its only network use is fetching a Whisper model and its tokenizer from Hugging Face after you tap Download — through multiple requests for file listings, repository information, and the metadata and contents of the files fetched. They name the public files being fetched and carry nothing you recorded, wrote or signed in as. Like any web request they do show Hugging Face your address, which is why the store's privacy label names three kinds of data, all of them from that download. The privacy policy says which, and why.

One permission, once

The microphone, the first time you record, and that is the only thing iOS asks you about — transcribing needs no permission of its own. Not your location, contacts, photos, camera, notifications or Bluetooth, and no request to track you, because nothing here is doing any.

The privacy policy goes through every file the app writes and every request it can make, and then the things that are not the app's doing at all: iOS fetching Apple's speech languages, what Apple reports to a developer through the App Store and TestFlight, and what we have if you write to us.

Get it.

The Plain Said app icon: a cream wave settling into three cream lines on a hot pink tile.

iPhone

Download on the App Store

Or join the beta on TestFlight.

It needs iOS 26.0 or later on an iPhone, and runs on an iPad as any iPhone app does, portrait only. The app is a few megabytes; a Whisper model, if you want one, is much larger and comes down separately.

Questions, bugs, or a transcript that came out wrong in an interesting way: stray@grunt.be. Support says what to put in the message so the answer is useful first time.