How Mermer works

They stay on your iPhone. Mermer has no networking code of its own, no account and no server, so there is nowhere for a recording to be uploaded to.

Transcription happens on the device too, using Apple's on-device speech model. Your audio is never sent away to be transcribed — not to us, not to Apple, not to anyone. Full detail in the privacy policy.

That's Apple's speech model. It's a component of iOS rather than part of Mermer, and iOS downloads it on demand the first time an app asks for on-device transcription. It is shared by every app on your phone that uses it, so it only ever downloads once.

Nothing is uploaded during that download. It travels from Apple to your phone, and no audio, text or information about you goes the other way. Mermer shows you the progress rather than hiding it, and you can carry on recording while it finishes — the transcript arrives once the model is installed.

Pull down on the list to reveal the search field and type a phrase you remember from inside the recording. Mermer searches the transcripts, not just the titles, and shows you the matching line so you can tell which recording is the right one.

You can also rename a recording — open it and tap the rename button in the toolbar — which makes it easier to spot in the list later.

Yes. Start the recording, lock the phone or put it in your pocket, and it keeps going. Mermer also survives interruptions like a phone call: when the interruption ends, recording resumes.

Stop the recording when you're done and the transcript is produced straight afterwards.

Two things run over your recording. Apple's speech model produces the words and their timings. A separate pair of small models, included inside the app, listens for the characteristics of each voice and works out how many people are talking and when each one starts and stops.

Mermer then matches the words to the voices word by word, so a fast exchange doesn't end up attributed to one person. Both steps run on your device, and the speaker models need no download and no network access at all.

Yes — tap the speaker's label in the transcript and type a name. It applies to every turn by that speaker in that recording, and to anything you export or share.

Clearing the name puts the numbered label back.

When you name a speaker, you can also choose to remember that voice. From then on Mermer recognises that person in future recordings and labels their turns with their name automatically, instead of calling them "Speaker 2" every time.

What gets saved is a set of numbers derived from their voice — not a recording. It cannot be played back or turned into audio. It lives on your device only.

Manage or remove them on the Remembered Voices screen, reachable from the toolbar on the recordings list. You can rename or delete any voice there, and deleting is permanent and immediate.

Remembering needs a few seconds of that person speaking in the recording. If there isn't enough of their voice to work from, Mermer says so rather than saving something unreliable.

Speaker separation is very good but not perfect, particularly when people talk over each other or a voice appears only briefly.

Two fixes:

• Rename the speaker in that transcript — the quickest correction, and it applies everywhere in that recording.
• Delete and re-save the remembered voice if someone is being recognised as the wrong person. Enrol it again from a recording where that person speaks clearly and for a good stretch.

Recording quality matters more than anything else here: the further away the phone, and the noisier the room, the harder the job.

Recordings made before speaker identification existed keep their original transcripts. Open one and use Detect Speakers — Mermer rebuilds that transcript with speaker turns.

On-device transcription is accurate but not infallible. Names, jargon, crosstalk and background noise are where it slips.

The audio is always the original — nothing is re-encoded or trimmed to make the transcript — so you can play the recording back at the point in question. Turning timestamps on makes it much easier to find that point.

Mermer transcribes in your iPhone's current language. If your language has no on-device model available, Mermer tells you so plainly instead of failing silently, and recording still works.

Open a recording and use the share button. You can share the audio file or the transcript, and the transcript goes out as formatted text with speaker names — straight into Mail, Messages, Notes, Files or anywhere else iOS can share.

Right after deleting, an Undo button appears at the bottom of the list for a few seconds. Tap it and the recording and its transcript come straight back.

Once that has gone, the deletion is permanent — Mermer has no server holding a copy. If a recording matters, share a copy of it somewhere.

There's no catch. No subscription, no in-app purchases, no ads, no locked features, and no limit on how much you record or transcribe. There is no payment code in the app whatsoever.

There's no data being collected in lieu of payment either: no account, no server, no analytics. Transcription is free to run because it happens on your phone rather than on rented computers.

Yes. Every control is labelled for VoiceOver, and the app is checked automatically against unlabelled controls so that can't quietly regress.

Mermer supports every Dynamic Type size including the accessibility sizes — transcripts scale all the way up, while the record bar stays usable.

Mermer requires iOS 26 or later, because the on-device transcription it is built around arrived in iOS 26.

It's an iPhone app. It will run on an iPad, but the layout isn't designed for the bigger screen yet.

Yes — Mermer is open source under the GPL. Everything the privacy policy claims about what the app does and doesn't send anywhere can be checked rather than taken on trust.

The Mermer name and its artwork are not part of that licence.