Live captions

Live captions are text of what’s being said, generated and shown on screen in real time, a line or two at a time and in step with the speech, so a viewer can read a conversation they can’t fully hear. In a meeting they come from one of three places: the video platform (Zoom, Microsoft Teams, Google Meet, Webex), the operating system (Live Captions on Windows 11 and macOS), or a human captioner. Automatic captions are produced by streaming speech recognition, the same technology behind a live transcript.

Captions are an accessibility feature first. For people who are deaf or hard of hearing, they’re how you take part in a call, not a convenience. They’re also an everyday tool for everyone else: a noisy room, a bad microphone, an unfamiliar accent, a meeting in your second language.

One idea explains both what captions are good for and where they stop: captions replace your ears, not your memory. They’re built to be read once, as the words are spoken, and then to get out of the way. So they’re excellent when you’re watching and can’t hear. They’re close to useless when you could hear fine but weren’t paying attention for the last few minutes.

Where meeting captions come from

Source	Examples	What it can hear	Who decides whether it’s available	Where the speech is processed
The meeting platform	Captions in Zoom, Teams, Google Meet, Webex	That platform’s meetings only	You turn them on for yourself, if the host’s plan, admin settings and meeting settings offer them	The platform’s servers
The operating system	Live Captions in Windows 11 and macOS	Any app the computer is playing: a meeting app, a browser tab, a video	You	On the device
A human captioner (CART)	A professional captioner listening to the call and writing in real time	Whatever audio they’re given	Usually the organizer or an accessibility team, who book it in advance	A person

Platform captions are the most common. They’re usually one click once they’re offered, and some platforms add the speaker’s name to each line, because the platform knows whose microphone is live. Whether they’re offered at all can depend on the host’s plan, the meeting settings and your organization’s admin policy, and that varies by platform and changes between releases, so check your own account. They only work inside that platform. The per-platform details are in the Teams and Google Meet answers.

Operating-system captions caption whatever the computer is playing, in any app. Windows 11 added Live Captions in its 22H2 update (Win+Ctrl+L). At the time of writing macOS has Live Captions on Apple silicon Macs, in some languages and regions. Both process speech on the device. They work anywhere because the operating system can already hear every app, which is the same access a system audio capture tool asks permission for. Details and setup for each are in the Mac and Windows answers. Some browsers and phones have their own caption settings too, limited in the same way to the audio they can hear.

Human captioning is usually called CART (Communication Access Realtime Translation). A trained captioner writes what’s said as it’s said, often using stenography. Their text may appear in the meeting’s own caption area or on a separate web page, depending on the platform. It’s a paid service booked ahead of time, and for deaf and hard-of-hearing people in meetings that matter it’s still the gold standard. A person can handle names, jargon and crosstalk that automatic captions get wrong, and can mark who is speaking and what’s happening. Sign language interpretation is a separate service, for people whose first language is signed. Captions don’t replace it.

Why captions scroll away

Everything about how captions look follows from their one job, which is to stand in for hearing.

• They keep pace with speech. A caption line appears a second or two after the words are spoken and is replaced as soon as there’s more to show. The words can shift while you read them, because the recognizer revises its guess as more audio arrives. That’s interim transcription results showing through.
• They show a line or two, not the whole conversation. A caption area is small on purpose, so it doesn’t cover faces and shared screens. Some platforms put a scrollable transcript pane beside the captions. That’s a transcript: the same words in a different container.
• Every sentence gets equal weight. Captions have no idea which sentence was the decision and which was small talk. They show everything at the same size, in order, and then it’s gone.

That design is right for reading along. It’s wrong for catching up. If you looked away for three minutes, the sentences that mattered have already scrolled off. Even with a transcript pane, reading back three minutes of dialogue while new lines keep arriving takes longer than you have when someone has just asked for your view.

Captions vs subtitles, transcripts and summaries

These get mixed up constantly:

• Closed vs open captions. Closed captions can be switched on and off by each viewer. Open captions are burned into the picture and always visible. Meeting captions are closed captions: each person decides for their own screen.
• Captions vs subtitles. In American usage, captions are for people who can’t hear the audio, and good ones mark who’s speaking and describe sounds such as [laughter]. Subtitles are for people who can hear but don’t understand the language, so they’re a translation. In British usage “subtitles” usually covers both, and captions for deaf viewers are called subtitles for the deaf and hard of hearing (SDH). Automatic meeting captions mostly render speech and little else. Some platforms also offer translated captions, which are live subtitles in this sense. Do AI meeting notes work in other languages? covers that side.
• Captions vs a live transcript. Same words, different container. Captions are a display: read now, then gone. A transcript is a document, kept so you can scroll it, search it and quote it. Platforms often produce both from the same recognition stream.
• Captions vs a live summary. A summary doesn’t reproduce the words. It condenses them into what’s being discussed. A rolling summary re-condenses as the call goes on, so a glance tells you where things stand. Live transcript vs live summary covers the difference in full.
• Captions vs caption scraping. Caption scraping is what some browser-extension notetakers do: read the captions a platform is already drawing in your tab and keep them as a transcript. It takes something built to disappear and turns it into a record.

What captions fix, and what they don’t

The quickest way to know whether captions will help is to name what actually went wrong:

What went wrong	Do captions help?	Why
You couldn’t hear it: noise, a bad mic, crosstalk, an accent, hearing loss	Yes. This is their job.	The words are on screen as they’re said
You heard it but couldn’t process it in time: a second language, auditory processing	Partly	You get a second look at the last line, but it moves on at speaking pace
You looked away for a few minutes	No	The lines that mattered have scrolled off, and rereading minutes of dialogue takes too long
You were there but lost the thread	No	Every sentence gets equal weight, so nothing tells you which ones mattered
You need the exact words tomorrow	Only if a transcript was saved	Captions aren’t a record unless something keeps them

Automatic captions also make mistakes, and they make the most in exactly the conditions where you need them: crosstalk, names, jargon, poor microphones and accents the model has heard less of. Word error rate explains how that’s measured and why a headline accuracy figure says little about your meeting. For the second row, when the words arrive but understanding lags behind, keeping up with auditory processing disorder goes through what to pair captions with.

Captions and transparency

Turning on captions for yourself usually changes only your own screen. On several platforms the other participants aren’t notified, because captions are treated as a display setting, like turning up the volume. Operating-system captions run on your device and, at the time of writing, are built as a display rather than a recorder, so check your version if it matters.

The line worth holding onto: reading along isn’t recording, but keeping the text is. Once the words are saved, whether as a platform transcript, by an extension collecting captions, or by a notetaker, there is a record of what everyone said. Most platforms tell participants when their own transcription starts. A tool running on your computer may not tell anyone, so telling people is your job. How to tell participants you’re using an AI notetaker has wording that works, and the rules on recording vary by region.

Where a live summary fits

Captions and a live summary cover the two halves of keeping up: captions do the ears, a summary does the memory. They work well together.

Canary is built for the memory half. It’s a real-time, bot-free meeting summarizer. It captures your computer’s system audio (no bot in the call, no plugin) and shows a live, multi-resolution rolling summary, from what’s being said right now to the whole call, so you can catch up the instant your name is called. That’s the moment captions can’t help with: you’ve been away for a few minutes, and the lines you missed have scrolled off. See what did I miss in the meeting?

The limits:

• Canary doesn’t show captions. Its floating widget shows the summary, not the words. If you need to read what’s being said, keep captions on as well. Operating-system captions run alongside it in any app. After the call, Canary’s review view has the full transcript, which you can search and copy.
• It isn’t on-device. Operating-system captions transcribe on your computer. Canary streams audio to a cloud speech-recognition service. If “nothing leaves my computer” is a hard requirement, built-in captions meet it and Canary doesn’t. On-device transcription explains what that property does and doesn’t buy you.
• It’s not an accessibility accommodation. Canary uses the same kind of speech recognition as automatic captions, so it has trouble with the same bad audio, and a summary paraphrases rather than quoting. For a deaf or hard-of-hearing colleague, the right answer is captions, CART or an interpreter, arranged by the meeting’s organizer. A summary is a supplement, not access.

For how the platforms’ own features compare, see Canary vs Teams live transcription and Canary vs Google Meet. For the closest third-party tool built on captions, see Canary vs Tactiq.

Related

• Streaming transcription: the speech recognition that produces automatic captions.
• Caption scraping: how browser extensions turn a platform’s captions into a transcript.
• Live transcript vs live summary: the words versus the meaning.
• How do I keep up in meetings with auditory processing disorder?: pairing captions with a summary.
• How do I catch up after stepping away from a call?: the case captions don’t cover.