Guide · macOS · updated 2026-09-06
Live captions on a Mac do not translate. Here is what does.
The single fact that sends most people to this page: macOS Live Captions transcribe, they do not translate. Apple's own support page describes turning spoken audio into text in the same language, on Apple silicon Macs, with accuracy that varies — and Apple's Live Translation is a different feature living in Messages, FaceTime and Phone, not in your meeting client. So if you need captions in another language there are three routes, and they differ mainly in which audio they can hear. System captions hear every app but only transcribe. Browser extensions and web tools translate, and hear exactly one browser tab — not the Zoom or Teams desktop client, not a podcast app, not a call on speaker. An app built on macOS system audio capture hears anything the Mac plays and can show both languages, at the cost of one permission and, usually, your own API keys. Below: what each route does not do, and the measured delay for the third one, because translated captions are always further behind the speaker than plain ones.
Route one: the system feature
- System Settings → Accessibility → Live Captions.
- Turn it on. By default it captions audio from every app; it can be limited to FaceTime.
- Read what it does not do: same language only, no translation.
The conditions are on Apple's page: Apple silicon only, and not available in every language, country or region — Get captions of spoken and computer audio on Mac. Windows has an equivalent that also transcribes rather than translates: Use live captions to better understand audio.
This is the right route if your problem is hearing rather than language: it costs nothing, installs nothing, and works everywhere on the machine.
Route two: a browser extension or web tool
These do translate, and they install in a minute. Their boundary is equally sharp: they hear one browser tab. Everything that matters in a working day — the Zoom desktop client, the Teams desktop client, a podcast app, a phone call on speaker — is outside what they can hear. If you join meetings in the browser they work; if your team uses the desktop apps they do not, and no setting changes that.
We have not measured any of them, so this page does not quote latency figures for them. Untested is written as untested.
Route three: capture the system audio
macOS exposes system audio capture through ScreenCaptureKit, so an app can caption whatever the Mac is playing with no virtual audio driver, no change to your output device, and no joining the meeting as a participant. You grant Screen & System Audio Recording once. The mechanics, including the failure mode where the stream starts and no audio ever arrives, are in recording system audio on a Mac.
Measured on this route, in a real 31-minute English meeting on 2026-09-03: English words on screen 0.71 s p50, first character of the Chinese translation 2.81 s, end of sentence to settled translation 0.81 s, and zero rewrites per sentence. The gap between 0.71 and 2.81 is the honest cost of translation: transcription can stream word by word, translation has to wait for enough of a clause to be worth translating. Anyone quoting one number for "live translated captions" is quoting the easy half. The full set, across three engine chains, is on the measured data page.
The app those numbers come from, Voice Translator (formerly Meeting Translator), is documented in Chinese: it is built for native Chinese speakers in English meetings, and its second job — handing you an English sentence when it is your turn — only makes sense for that reader.
Three routes, one table
| Dimension | macOS Live Captions | Browser tool | System audio capture |
|---|---|---|---|
| Which audio it hears | Every app, or FaceTime only | One browser tab | Everything the Mac plays |
| Translates | No — transcribes only | Yes | Yes, both languages on screen |
| Latency | Untested | Untested | 0.71 s transcript, 2.81 s translation (2026-09-03, real meeting) |
| Requirements | Apple silicon Mac | An extension or a web page | macOS 26, Apple silicon, one permission, your own API keys |
| Where the audio goes | Stays on the Mac | Depends on the tool | On-device chain stays local; cloud engines are named next to the button |
| Cost | Free, built in | Varies | App plus per-minute engine cost |
Related: how the capture actually works · convert the captions you saved · about Voice Translator
Questions people actually ask
Do macOS Live Captions translate?+
No. Live Captions turns spoken audio into text in the same language; Apple's support page describes transcription, not translation, and notes that accuracy varies and the feature should not be relied on in high-risk situations. It is also limited to Apple silicon Macs and is not available in every language, country or region. Apple's Live Translation is a separate feature and lives in Messages, FaceTime and Phone — it does not caption a Zoom call.
What is the difference between live captions and live caption translation?+
Live captions show you what was said in the language it was said in. Live caption translation shows it in another language, which needs a second step — machine translation — and that step is where the extra delay lives. On our own measurements the English words appear at 0.71 s p50 while the first character of the translation arrives at 2.81 s: the translation is roughly four times further behind the speaker than the transcript is.
Which audio can each route actually hear?+
System captions hear every app, or FaceTime only, depending on the setting. Browser extensions and web tools hear one browser tab — not the Zoom or Teams desktop client, not a podcast app, not a phone call on speaker. Something built on system audio capture hears anything the Mac plays. Pick by where your audio actually comes from, not by which tool has the nicest page.
Do I need a virtual audio device?+
Not for the system feature, and not for an app built on ScreenCaptureKit, which macOS provides for capturing system audio without a driver and without changing your input or output device. You grant Screen and System Audio Recording once. A virtual audio device is only needed when the sound must be fed to an app that expects a microphone.