Live captions for lectures,
when you cannot ask for a repeat.

Missed terms compound

A conversation tolerates a lost word. A lecture does not, because the definition you missed at ten past is what the next twenty minutes are built on. By the time you notice, the thing you needed is four slides back and nobody is going to return to it for you.

It is worse because you are also typing. Notes and listening compete for the same attention, and the moment you look down to finish a sentence is reliably the moment something important is said. The overlay stays above your notes window and takes no clicks, so the last line said is on screen while you are still writing down the one before it.

And if the lecture is not in your first language, all of that happens at the speed of somebody else's accent.

14-day refund, no questions asked.

Pick a slower model than you would for a call

The English models trade latency against accuracy, and a lecture wants the opposite end of that dial from a conversation. In a call you want the 160 ms model because you are about to reply and a late caption is a useless one. In a lecture you are not replying to anybody, so the extra second costs you nothing and buys accuracy.

Nemotron at 1120 or 2240 ms is the one to use here. It punctuates and capitalises as it goes, which matters a great deal when you are reading a definition rather than a sentence someone just said to you. Switch it in the menu bar, per lecture if you like.

Lecture questions

Does it work with Panopto, Echo360, Canvas and YouTube?

All of them. It captures the audio an app produces rather than integrating with a platform, so a lecture in Panopto, an Echo360 capture, a Zoom seminar, a YouTube recording and a video embedded in Canvas are all the same thing to it: sound coming out of your Mac.

Will it get the technical terms right?

Often not, and no speech recogniser will. Universities say as much in their own guidance about automatic captions: accented speech and specialist vocabulary are where accuracy drops. This helps with the sentence you half-heard, not with a term the model has never seen. Expect to correct names, notation and jargon yourself.

Can I keep the captions to revise from?

No. Nothing is written to disk here, ever. For revision, use your institution's recording and whatever transcript it produces, which is a better tool for that job and usually included in your fees.

What if my university has not enabled captions?

Then this works anyway, because it never asks your university anything. Whether platform captions exist depends on an institutional setting and a confidence threshold, and neither is something you can change from your seat.

Does it work in a lecture hall?

No, and this is the one thing to be clear about before buying. It captions audio your Mac is playing and never opens the microphone, so a lecturer speaking in a room is not something it can hear. It is for lectures you stream, join or play back.

Which model should I use for a lecture?

One of the slower ones. In conversation you want the 160 ms model because you are about to reply; in a lecture you are not replying to anyone, so trade latency for accuracy and use Nemotron at 1120 or 2240 ms, which also punctuates and capitalises.