Subtitles

Subtitles · Guides · Real-time subtitles on a Mac

Real-time subtitles for anything your Mac plays

Six ways to put words on screen while someone is still speaking, ordered by what they cost you, which means the free ones come first. Then the part almost nobody explains: how a Mac captures the sound it is making, why that used to need a fake audio device, and what changed in macOS 14.2.

14-day refund, no questions asked

Buy for $9 Try it in your browser

What "real-time" rules out

Most of what is written about transcribing on a Mac is about files. You drop in a recording, wait, and get a transcript or a subtitle track. That route is more accurate than anything on this page, because a model that can see the whole sentence, and the sentence after it, will always beat one deciding a word at a time.

It is also useless while a meeting is happening. Live captioning trades that accuracy for the only thing that matters during a call, which is that the words arrive in time to be read. If what you have is a recording and time to spare, stop here and run it through a file transcriber; you will get a better result. Everything below is for the case where the audio is happening now.

Six routes, cheapest first

  • The app's own captions. Free. Zoom, Teams, Google Meet, Slack huddles and YouTube all caption themselves, at no cost, in more languages than macOS does. They stop at the edge of their own window, several of them need the host or an admin to allow it, and none of them help with the video someone links to halfway through. Try this first anyway: when it is enough, it is free.
  • Apple's Live Captions. Free, built in. On an Apple silicon Mac, under System Settings, Accessibility, Live Captions. It captions across apps rather than inside one, which is the right shape. Apple's own documentation says it is not available in every language, country or region, and the captions arrive in a window that behaves like a window. The full comparison is its own page.
  • Your browser's live caption. Free, one browser. Chrome and the Chromium browsers can caption audio playing in a tab, on device, from a switch in their own settings. Genuinely good for video in a tab, and irrelevant to everything outside that browser.
  • A cloud captioning service. Subscription. The category that reaches languages nothing on-device covers, and the one that adds speaker labels and a searchable transcript afterwards. The audio leaves your Mac to get there, which for many people at work is the end of the conversation rather than a detail.
  • A local speech model you drive yourself. Free, and work. Whisper and its descendants run on a Mac and there are good open-source front ends. Making one caption live, across every app, in an overlay you can read while you work, is a project rather than a download, and the audio plumbing below is most of why.
  • A live captioning app. Paid, once. What this is: an overlay above every window, fed by whatever the Mac is playing, transcribed on the Neural Engine, in sixteen languages, with nothing written to disk. Worth it when the four free answers above have all run out, and not before.

How a Mac captures its own sound

A microphone is easy: macOS has always had an API for that, and asking permission for it is routine. System audio is the hard one, because for most of the Mac's life there was no supported way for one app to hear what another app was playing. Three eras, and which one a tool belongs to tells you most of what you need to know about it.

  1. 1

    The virtual audio device

    You install a driver that pretends to be a pair of speakers, set the Mac's output to it, and read it back as a microphone. It works, and it means your actual speakers are now silent unless you build a second device that splits the signal. This is the era of the setup guides with six screenshots in them.

  2. 2

    Screen recording, for the audio

    Later macOS let an app take system audio along with a screen recording. No driver, but you are granting an app the right to see your screen in order to hear it, and macOS reminds everyone in the room with a recording indicator.

  3. 3

    Core Audio process taps, macOS 14.2

    An app asks Core Audio to tap another process, or all of them, and gets the audio directly. No driver, nothing rerouted, your speakers keep working, and the permission you grant is audio recording rather than screen recording. This is the whole reason this app exists in the shape it does.

It also explains the one requirement that has no way around it. Process taps arrived in macOS 14.2 and only on Apple silicon, so 14.2 is the floor here and an Intel Mac cannot be raised to it. If you are on Intel, the first three routes above are your list.

Captioning your Mac, in three steps

  1. 1

    Install, and grant the one permission

    macOS asks to record system audio on first run. Miss that prompt and every buffer arrives full of silence with no error anywhere, which is the one failure worth knowing about in advance. The set-up guide covers spotting it in ten seconds.

  2. 2

    Point it at everything, or at one app

    All system audio is the default and needs nothing set. Narrowing it to one app is a menu away, and it takes that app's helper processes with it, which is what makes a call in a browser tab caption like a native one.

  3. 3

    Put the overlay where you want it

    Hold ⇧ and drag; where you leave it is remembered. It takes no clicks, so you work straight through it, and pointing at it makes it dissolve rather than move. The demo on the home page behaves exactly the same way, in your browser, before you buy anything.

Buy for $9

14-day refund, no questions asked.

What real-time costs in accuracy

The recogniser is a streaming model rather than one that waits for a thirty-second window, which is the difference between captions that track the audio and captions that arrive after the moment has passed. It runs at roughly a seventh of real time on the Neural Engine, so the machine is not working hard and the fans stay where they were.

English gets seven checkpoints rather than one, and they trade the two things that cannot both be maximised: the fastest starts producing text about 160 ms behind the speaker, the slowest is more accurate and further behind. Everything else runs from one multilingual model. That dial is the honest version of the accuracy question, and it is the reason this page does not claim to beat anything in a benchmark nobody has run.

What still goes wrong, in every tool on this page: names, jargon, acronyms, two people talking at once, and fast colloquial speech down a bad connection. Live captioning makes those readable often enough to be worth having, not reliably enough to be a transcript.

Questions about captioning a Mac

How do I caption system audio on a Mac?
Since macOS 14.2 an app can tap another process's audio directly through Core Audio, so all it needs from you is the audio recording permission. Before that you had to install a virtual audio device and route your output through it, which is why the older guides are full of screenshots.
Do I still need BlackHole or a loopback driver?
Not for this. Nothing is rerouted, your output device does not change, and your speakers keep playing while the captions run. A loopback driver is still the answer if you are on macOS 13 or older, or on an Intel Mac.
Is real-time as accurate as transcribing a file?
No, and nothing live is. A file transcriber sees the whole sentence before deciding what the first word was; a live one has to commit while the speaker is still talking. If you have a recording and no hurry, transcribe the file instead.
Does it work in a browser tab?
Yes. It captures an app and every helper process that app spawns, so a call or a video in a tab captions the same way a native app does. That also covers Electron apps, whose audio never comes from the process wearing their name.
Does it caption my own voice?
No. The microphone is never opened. It hears what your Mac plays, which is everyone except you, and that is a deliberate line rather than a missing feature.
How much delay is there?
The fastest English model starts producing text about 160 ms behind the speaker; the more accurate ones sit further back. It is a dial rather than a number, and which end of it you want depends on whether you are following a conversation or reading a lecture.
Can I save the transcript?
No. Nothing is written to disk, on purpose. Hold ⌥ to bring the last few captions back and scroll through them, but once they scroll off they are gone. If you need a record afterwards, you want a file transcriber or a notetaker, not this.

The same overlay, elsewhere

Videos with no subtitles

Subtitles over a film, a stream or a tab that never had a track.

Podcasts

Captions for the shows that never had a transcript.

Compatible apps

If it plays sound on your Mac, it captions. The list, and the exceptions.

vs Apple Live Captions

macOS has this built in and free. Where it stops being enough.

Every alternative, compared

Six ways to get captions on a Mac, and which one fits you.

What the app keeps

Nothing is written to disk, and after the model downloads there is no network call left.

Subtitles for the thing playing right now.

14-day refund, no questions asked

Buy for $9

One-time $9, every future update included · macOS 14.2 or later · Apple silicon
Unlimited live captions, for as long as you own the Mac. No minutes to buy, no account, no subscription.