Campus Captions Request access

Campus Captions

Rev. Beta — not for sale

Live captioning for events — wall · phones · stream

Cost / hour On the on-device engines. No account, no metering.
$0.00
Audio leaves the room No

Apple and Whisper transcribe on the booth machine itself.

Requires

  • 1 × Windows or Mac host
  • 1 × line-level feed from the desk
  • LAN reachable by wall + phones
  • Internet — optional

Outputs

  • Browser source, transparent
  • Phone page over LAN, QR to join
  • YouTube Live captions, pushed
  • .srt transcript on close
Private beta · adding venues slowly Request beta access

Speech recognition is a part you choose, not a vendor you hire.

Two of the four engines run entirely on the machine in your booth. No account, no API key, no per-minute billing, and after the first model download, no network at all. That is the whole argument, and it is checkable.

Deepgram is the only engine that bills, and the only one that sends audio anywhere. It is an option, not the floor.
Engine Cost / hr Runs on Audio leaves the room
Apple $0.00 macOS 26+ No
Whisper $0.00 Win · Mac · Linux No
Deepgram Metered, your account Any, online Yes
Rehearsal $0.00 Anywhere No audio at all

Rehearsal replays a script you write, with no microphone and no key — which is how you train a volunteer at two o'clock on a Tuesday without booking the room.

Wall outputBrowser source · transparent · any resolution

Rehearsal Roll-up · 3 lines Atkinson Hyperlegible

The wall view, replaying a written script through the rehearsal backend — the same mode you train a volunteer on, so nothing above is a recording of a real event. Captions roll up from the bottom; the newest line is the bright one, and the face is the reader-selectable one drawn for low-vision legibility. Drop it into ProPresenter, OBS or vMix as a layer over whatever is already on screen — font, size, position, alignment and line count all set from the URL.

Where the words goOne capture · three destinations

The wallProPresenter · OBS · vMix

A transparent browser source you add once as a web Prop, a browser source or a title. It can also drive ProPresenter directly, with no web layer at all.

It never shows a spinner, an error or a placeholder on a screen an audience can see. When something breaks the bar goes empty — not red.

Every phoneLAN · QR to join

Put the QR page on screen before doors. Anyone who scans it reads along on their own phone, at their own text size, in their own colour, with scrollback — so arriving late or glancing away costs nothing.

No app, no account, no sign-in. It is a web page. Each reader can pick a language if you have turned translation on.

The streamYouTube Live · .srt

Captions push into YouTube Live as the event runs. Afterwards you get a timed .srt of every committed line, so the archive video carries real captions instead of auto-generated ones.

The booth, most nights, is one student worker running three other things.

First run

A setup wizard — microphone, engine, event title — then a guided tour that points at each control on the real page and says what it is for. Every run after that opens straight on the status page.

Before doors

A preflight check reads the network, the audio and the config and returns plain-language findings, each with what to do about it, and one verdict: ready, ready with warnings, or not ready.

During

One page shows whether audio is arriving, whether text is coming back, how many viewers are connected, and whether the engine stream is open. Most of the job is watching one green dot.

There is a Bitfocus Companion module too, so the caption controls sit on the same surface as the switcher rather than in a browser tab nobody can reach.

Proper nouns

You give it a list — people, places, programmes, the words peculiar to your campus — and the recognizer is primed with them before anyone speaks. There is a blocklist too, for words you never want appearing eight feet tall behind a speaker.

Known limitationsStated before you ask for them

Installers unsigned

Neither build is code-signed, so each shows a one-time operating-system warning the first time it runs on a given machine. Installation is not frictionless and we will not pretend otherwise.

Accuracy varies

It depends on the engine, how clean the feed off the desk is, and whether you supplied a keyterm list. No accuracy figure has been measured, so none is quoted here. Good enough that a room follows along; not good enough to be the official record of what was said.

Remote viewers

Reaching people outside the building needs a tunnel you set up. On the local network it just works.

No price

It is not for sale and no price has been set. Inventing one before the beta tells us what this is worth would be guessing at your expense.

Automatic captions are not a legal accommodation. Where someone has a documented accommodation requiring verbatim captioning, that remains a human CART writer's job.

Your disability services office makes that call, not us. Campus Captions is for the much larger set of events that today have no captions at all.

Request beta accessWe add venues slowly, on purpose

Tell us about your room. We are taking on a small number of sites so each one gets actual attention, and we will be straight with you about whether it is a fit.

Used to contact you about the beta. Nothing else, and nobody else.

Notes

Does it work with ProPresenter?

Two ways. Add the wall view as a web Prop and lay it over your slides — that is the usual setup. Or send captions into ProPresenter directly, with no web layer, if your operator would rather drive it from inside ProPresenter.

Do we need internet?

Not for the captions, on either on-device engine. Audio never leaves the computer and nothing is billed.

You need the local network so the wall and phones can reach the relay. Internet is only for the optional extras: the Deepgram engine, translation, pushing captions to YouTube Live, or letting people outside the building connect.

What do we need to buy?

Most likely nothing. One Windows or Mac computer — the one already in the booth is usually fine — and a feed from the desk you already run. If there is no spare output, a cheap USB interface covers it.

How accurate is it?

It depends on three things you control: the engine, how clean the feed is, and whether you gave it your keyterm list. A clean lectern mic plus a list of your proper nouns is a different experience from a room mic and no list.

No figure is quoted because none has been measured.

Can the audience read in another language?

Yes, if you turn it on. Every completed line is translated into the languages you offer, and each person picks theirs on their own phone. A phone joining halfway through and choosing Spanish replays the Spanish history, not the English. Only languages with a working translator behind them are ever shown.

What does the beta cost?

Nothing, and no price is set for afterwards. If you run the Deepgram engine you pay Deepgram directly for your own usage; the on-device engines avoid even that.