Offline transcription on iPhone: what runs on the device, and how to set it up

What “offline” actually covers

A transcription app is offline when the model that turns speech into text is stored and run on the phone. The audio stays where it was recorded, and the result appears without a network round trip. Two things still need a connection: downloading the model the first time, and any optional service you add later, such as an AI provider or cloud sync.

It helps to separate three questions: where the audio is stored, where transcription runs, and what else can leave the phone. A good answer names all three.

Option 1: Voice Memos, built into iOS

Apple’s Voice Memos app is free and already on your iPhone. Apple’s iPhone User Guide describes viewing a transcription of a recording on supported models and languages, and Apple lists the requirements there. If you want a transcript of one recording at a time, it is the simplest choice and worth trying first. Our Voice Memos comparison covers where it is enough. The guide lists iPhone 12 or later and ten languages, not in every region. It does not say where the transcription is processed; if that matters to you, check Apple’s privacy information.

Option 2: Mnemo, with a choice of on-device engines

Mnemo only accepts transcription engines that run on the device; an engine that could run off the device is refused when it is registered, so a hosted speech service cannot be plugged in. You choose between these engines:

• Apple Speech, on the device. On iOS 26 Mnemo uses Apple’s newer speech engine; on iOS 18 to 25 it uses Apple’s previous on-device recognizer. The available languages are decided by iOS on your device.
• Whisper Small, free, a 491 MB download. On devices where Small is not supported, such as iPhone 11 and some iPads, Mnemo offers Whisper Base instead, a 151 MB download that works best in English.
• With Pro: Whisper Large v3 Turbo (631 MB) and Parakeet v3 (632 MB), which covers 25 European languages.

You can lock a language or let Mnemo detect it, and re-transcribe a recording later with another language or model. The new transcript is a linked copy; the original stays intact.

Set it up in Mnemo, step by step

1. Open the speech model manager in Settings and download the free model. Mnemo shows the size and lets you choose Wi‑Fi or mobile data.
2. Check the processing setting. Out of the box Mnemo processes on the device only; Privacy Mode additionally blocks every optional network route.
3. Choose the transcription language, or leave automatic detection on.
4. Record a Meeting, Interview, Quick note or Dictation, or open an existing audio file in Mnemo from Voice Memos, Files or Mail.
5. Search the transcript, correct it without touching the raw version, and export to Markdown, TXT, JSON, CSV tasks, SRT or WebVTT, or to PDF through the share sheet.

What still uses the network, and only if you choose it

Four routes can reach the network in Mnemo: model downloads you confirm, which send no recording or text; an AI key you connect for Ask or Skills, which sends transcript text to that provider when you run them; optional iCloud sync of transcripts, never audio; and webhooks you configure. Privacy Mode blocks them. There is no Mnemo account and no analytics SDK. The privacy boundary page lists each route.

Limits

• No accuracy or speed figures are published for any engine in Mnemo. Try a short recording in your language and accent first.
• Whisper Base is best in English and less accurate in other languages, according to its catalogue entry.
• The first model download needs a connection.
• Mnemo is for iPhone and iPad with iOS 18 or later, and is not yet on the App Store.

Straight answers

Sources (accessed 28 September 2026)

• Apple Support: iPhone User Guide, view a transcription in Voice Memos