Compare
Best offline speech-to-text for Mac, including the catch
Every app on this page runs recognition with no connection. Every one of them also needs the internet once, to download its model. That is the catch, and it is the only one.
Last updated
Why offline is a separate question from privacy
They overlap — an app that runs locally is both private and offline-capable — but people arrive here for different reasons, and the requirements differ.
A privacy buyer wants to know the audio is not transmitted. An offline buyer wants the app to keep working when there is no signal. The second is a stricter test in one respect: an app can be privacy-respecting and still refuse to start without a licence check. Ask specifically about that.
What "works offline" should mean
- Recognition runs with the network off.
- The app launches with the network off.
- The licence does not need to phone home on every launch.
- Your vocabulary, snippets and settings are local.
- The history is local and readable offline.
The model download
Every local speech app needs its model, and models are hundreds of megabytes to a couple of gigabytes. They are downloaded once, on first run or when you change model, and reused for ever after.
The apps
| App | Offline dictation | Offline transcription of files | Notes |
|---|---|---|---|
| VV | Yes | No | Model downloaded and verified on first run; Apple Silicon (M1 or later) |
| VoiceInk | Yes | Check current release | Open source; Apple Silicon, macOS 14.4+ |
| Superwhisper | Yes | Yes | Their FAQ: offline models run best on Apple Silicon |
| MacWhisper | Yes | Yes | Built for files as well as dictation |
| Apple Dictation | On Apple silicon for supported languages | No | Free; check the Keyboard settings panel for your language |
| Wispr Flow | No | No | Cloud transcription by design |
Read on vendor sites on 11 September 2026; sources at the foot of this page.
VV
Dictation only, no file transcription. The speech model is downloaded and checksum-verified on first run and then everything happens on your Mac: recognition, vocabulary, snippets and history. Free is 1,000 words a week with no sign-in, which is enough to prove it on your own machine before a trip.
VoiceInk
Local transcription, open source, one-off licence. Apple Silicon and macOS 14.4 or later.
Superwhisper
Says it works offline, so you can transcribe anytime
, and covers both dictation and files. Its own FAQ notes that offline models run well on Apple Silicon and that Intel Macs are better served by cloud models — worth knowing if offline is the requirement and you are on Intel.
MacWhisper
The right answer if what you need offline is turning recordings into documents. Local models, Pro €64 one-off.
Apple Dictation
Free and already there. On Apple silicon it can run on your Mac for supported languages — check the line under the toggle in Keyboard settings, because it varies. Try this before buying anything.
Where offline dictation genuinely earns its place
- Flights. The single most common reason people buy. Long-haul is the best writing time many people get.
- Trains. Mobile data through cuttings and tunnels is worse than no data, because cloud apps keep trying.
- Basements, hospitals, older buildings. Thick walls beat Wi-Fi more often than anyone plans for.
- Site visits and fieldwork. Surveys, inspections, farm and construction work.
- Hotel and conference Wi-Fi. Technically present, practically useless.
- Air-gapped or restricted machines. Where cloud dictation is not merely inconvenient but forbidden.
Where these facts came from
Anything stated about another product was read on that product's own site or documentation on the date shown, and is re-checked at least quarterly. If something below has changed, tell us and we will correct it.
- Superwhisper says it "works offline, so you can transcribe anytime," and its FAQ adds that "Intel Macs work best with Cloud models. Offline models only run really well on Apple Silicon macs." source, checked .
- Superwhisper lists Mac, Windows and iOS. source, checked .
- VoiceInk says it "processes all voice transcription locally on your device" and that "your voice data never leaves your Mac," with optional cloud enhancement that handles transcribed text rather than audio. source, checked .
- VoiceInk requires an Apple Silicon Mac on macOS 14.4 or later; an iOS app is sold separately. source, checked .
- MacWhisper says it uses "local models to transcribe your files" and processes "sensitive content locally without data ever leaving your Mac," with optional connections to cloud AI providers. source, checked .
- MacWhisper covers both file transcription — audio, video, meetings and podcasts — and "Real-time dictation for messages, notes, and documents." source, checked .
- Apple tells Mac users to check Keyboard settings to see "whether your voice inputs and transcripts for general text Dictation … are processed on your device and not sent to Siri servers," and notes this varies by language and region. source, checked .
Questions
Does offline speech-to-text need the internet at all?
Once, for the model download, and afterwards only for things unrelated to recognition: updates, a licence check, or an optional cloud rewrite if you turn one on. Recognition itself needs nothing.
Is offline speech-to-text less accurate?
On clean audio from a decent microphone in a quiet room, the gap with large cloud models is small. In noisy conditions or with a strong accent in a second language, the big cloud models still lead. Test on your own voice rather than trusting either claim.
Does VV work on a plane?
Yes, once the model has been downloaded. Install and dictate one sentence before you travel — that is the only step that needs a connection. Free dictation also needs no sign-in, so there is no activation to fail in the air.
How much disk space does an offline model need?
A few hundred megabytes to a couple of gigabytes depending on the model and the accuracy you want. Downloaded once and reused.
Can I transcribe an audio file offline?
Not with a dictation app — VV takes live microphone input only. MacWhisper and Superwhisper both transcribe files with local models.