In short
On a Mac running macOS 26, Apple's on-device transcriber, SpeechTranscriber, supports ten languages: English, French, German, Italian, Spanish, Portuguese, Japanese, Korean, Chinese and Cantonese. We checked this on 25 September 2026 by asking macOS 26.6 for the list, which is what the operating system reports to apps.
Most of the ten come in several regional variants.
This guide gives the full list, explains why dictation on the same Mac covers more languages, and what to do if your meetings are in a language outside the ten.
The ten languages and their variants
Apple's reference documentation does not print a fixed list. Instead, apps ask macOS at run time through SpeechTranscriber.supportedLocales, which Apple describes as "the locales that the transcriber can transcribe into, including locales that may not be installed but are downloadable". This is what it returned on an Apple silicon Mac running macOS 26.6.2 in September 2026:
| Language | Variants (as reported by macOS) |
|---|---|
| English | Australia, Canada, India, Ireland, New Zealand, Singapore, South Africa, United Kingdom, United States |
| Spanish | Chile, Mexico, Spain, United States |
| French | Belgium, Canada, France, Switzerland |
| German | Austria, Germany, Switzerland |
| Chinese | China mainland, Hong Kong, Taiwan |
| Portuguese | Brazil, Portugal |
| Italian | Italy, Switzerland |
| Japanese | Japan |
| Korean | Korea |
| Cantonese | China mainland |
That is 30 locales across ten languages. Apple said at WWDC25 that the transcriber "can currently transcribe these languages, with more to come", so the list may grow in later releases. If you are reading this after a macOS update, treat it as a floor rather than a ceiling.
A variant matters more than it looks. The English models for India, the United Kingdom and the United States are separate locales, and choosing the one closest to how people in the meeting speak may give a better result.
Why dictation covers more languages
macOS dictation, the feature you start from the keyboard to type by voice, supports far more languages than the list above. It uses older speech models, the ones Apple also exposes to apps through DictationTranscriber and the on-device mode of SFSpeechRecognizer. On the same Mac, DictationTranscriber reported 54 locales.
The difference is the model. SpeechTranscriber uses Apple's newer model, which Apple describes as "good for long-form and distant audio, such as lectures, meetings, and conversations". Dictation's models were built for someone speaking into their own device, in short bursts. Apple's documentation for the older recogniser notes that on-device requests "won't be as accurate" as server ones, and that some languages are only available over the network.
Apple's Mac User Guide says "Dictation is not available in all languages or regions, and features may vary", and points to Apple's feature availability page for the current list. If you want to know which languages your Mac dictates, check there or in System Settings > Keyboard, where Apple says you can also see whether dictation is processed on the device.
Apple's SpeechAnalyzer compares the two APIs in more detail.
Each language is downloaded once
The models are not all on your Mac from the start. According to Apple's AssetInventory documentation, they are "downloaded from Apple's servers and managed by the system", kept once downloaded, updated automatically and shared between apps.
So the first time any app asks to transcribe, say, Japanese, macOS fetches the Japanese model. That needs a connection. After that, transcribing Japanese works offline. If you plan to transcribe with no network, use the language once while you are online. If transcription is not starting at all, speech recognition not working on a Mac goes through what to check.
If your meetings are in another language
If you meet in a language outside the ten, you have a few options, none of them perfect:
- Use a tool with its own model. Some transcription apps ship open-source speech models covering many more languages. Check whether they run on the device or upload the audio.
- Use a cloud transcription service. Cloud services often cover the most languages. The audio leaves your Mac and goes to the provider, which then becomes part of your data map.
- Wait for Apple. Apple has said more languages are coming to
SpeechTranscriber.
Meetings that switch between languages are harder still. A transcriber is set to one language at a time, so a sentence in another language may come out as the nearest-sounding words in the chosen one.
How Notey handles languages
Notey's live transcription uses Apple's on-device speech recognition, so it covers the ten languages listed above, on Macs with Apple silicon running macOS 26 or later. Recording works in any language; transcription depends on the list.
After an English meeting, Notey reads the recording again on your Mac and keeps the words the recognisers agree on. That step is English only; meetings in the other nine languages keep their live transcript. Why Notey re-transcribes a meeting after it ends explains it, and on-device transcription explained covers what "on-device" means in the first place.
Frequently asked questions
How many languages can a Mac transcribe on the device?
Apple's SpeechTranscriber, the newer on-device transcriber in macOS 26, reported ten languages on macOS 26.6 in September 2026, with 30 regional variants between them. System dictation covers more languages using older models.
Can a Mac transcribe a meeting in Dutch, Arabic or Russian on the device?
Not with SpeechTranscriber on macOS 26.6, which did not list those languages in September 2026. Apple says more languages are coming, so check again on newer versions of macOS.
Does Notey transcribe languages other than English?
Yes. Notey's live transcription covers the ten languages Apple's on-device recogniser supports. The improved transcript it makes after a meeting is English only.
Why does the first transcription in a new language need the internet?
macOS downloads the speech model for each language from Apple the first time an app needs it. After that the model stays on the Mac and transcription works offline.