Every live event eventually gets the same request: can we caption this? The answer depends far less on the captioning itself than on how the text reaches your switcher. Here are the four routes that actually work in a booth, and what each one costs you in flexibility.
1. Alpha-channel NDI — the clean route
If your production network already carries NDI, this is the one to use. The captions arrive as a video source where everything except the text is transparent, and the switcher composites it over the program feed exactly like a lower-third graphic.
- OBS Studio — add a NDI Source (via the DistroAV/obs-ndi plugin), tick the alpha/BGRA format, drag it to the top of your scene. If the transparent areas come in solid black instead of see-through, open the source’s Properties and set Latency Mode to Normal (safe) — Low or Lowest latency modes drop the alpha channel entirely.
- vMix — Add Input → NDI/Desktop Capture, then use it as an overlay channel rather than a full input, so it keys over whatever is live.
- TriCaster — bring it in as a network source and assign it to a DSK.
Because the text lives on its own layer, you can restyle, reposition or drop it mid-show without re-encoding anything. Nothing is baked into the recording either. We wrote up the mechanics separately in what alpha-channel NDI captioning is.
2. Chroma-key HDMI — when the venue has no NDI
Plenty of rooms run a hardware switcher and no production LAN. A Blackmagic ATEM, for instance, has no NDI input at all. The fallback is a full-screen HDMI output with a solid key colour behind the caption text, keyed out downstream on an upstream or downstream keyer.
You lose nothing visually. You do lose a display output on the caption machine, and you have to agree the key colour with whoever built the show file. Do that before doors, not during.
3. The platform’s own caption track — for online events
Zoom, Teams, Meet and Webex all render captions themselves. The catch is that they render one language at a time, chosen per viewer from a short list. For a single-language webinar that is enough and it is free. For anything multilingual it is a dead end.
SCAPTION handles online events by listening to your computer’s own audio, so it works the same way on every platform and nothing needs installing on the call. There is no bot joining the meeting — which also means there is nothing for a platform to block.
4. Attendee phones — the one that scales
The route most AV teams underestimate: put a QR code on the venue screen and let people read along on their own phone, each in whichever language they picked. It costs you no extra screen real estate, no hardware, and no seat-by-seat receivers.
This is what Scaption Access does — 50 languages at once, no app, no account, and no charge per attendee or per language. In our own load testing a single room held 5,000 concurrent phones; the size we recommend selling is 500–1,000, because that is where delivery stays comfortably inside the latency target rather than merely arriving.
Which one should you pick?
| Situation | Route |
|---|---|
| Production LAN with NDI | Alpha NDI |
| Hardware switcher, no NDI | Chroma-key HDMI |
| Single-language webinar | The platform’s own captions |
| Multilingual audience, any venue | Attendee phones, plus one of the above on the big screen |
Most real events end up running two at once: the venue screen carries the floor language, and the phones carry everything else.
You can try the whole thing on a real event free — three hours, valid 14 days, no card required. Download it for Windows and point it at a mic.