Agencies weighing live captions often frame it as AI captioning vs. human captioning, as if they must pick one. In practice it is rarely either-or. This guide explains what drives automatic captioning accuracy, how the costs are structured, and when each approach fits a public meeting.
Two ways to caption a live meeting
Human captioning for live events is usually CART, short for Communication Access Realtime Translation. A trained stenographer listens to the meeting and writes captions in real time, in the room or remotely. A good CART writer can prepare names and terms in advance, handle cross-talk and make judgment calls about what was said.
AI captioning, also called automatic or machine captioning, uses speech recognition to turn meeting audio into text. With SentraCaptions, the agency adds one more destination to the encoder it already streams with, over RTMP or SRT. Captions appear about one second after speech, and the sentence in progress updates live.
Both put text on a screen. They differ in how they get there, how they are scheduled and paid for, and how they handle the hard parts of a meeting.
Automatic captioning accuracy: what actually affects it
Be cautious with any single accuracy percentage, from any vendor. Accuracy depends heavily on the meeting itself, and a number measured on clean recorded audio says little about a council chamber during public comment. Focus instead on the factors you can control.
- Audio quality. This matters most. Captions are made from the same audio your stream carries. Speakers close to their microphones, with little echo or background noise, get better captions. See chamber audio tips for better captions.
- Local names. Council member names, streets, neighborhoods and program names are where general speech recognition stumbles. A glossary of local names helps captions spell them right. See building a caption glossary.
- Cross-talk. When two people talk at once, every captioner, human or machine, has a harder time.
- The language being spoken. SentraCaptions recognizes English speech, and Spanish speakers at public comment are recognized too.
Every SentraCaptions screen labels the captions as automatic and says they may contain errors. That label matters: residents should know what they are reading and how much weight to give it.
How the costs differ
We won't put dollar figures here. Prices vary, and we don't publish ours. What we can describe is how the cost is structured, which is often the bigger difference for a clerk's budget.
Human captioning
- Usually booked per meeting or per hour, often with a minimum.
- Has to be arranged ahead of time, which can be hard for special meetings or meetings that run late.
- Captions in other languages generally mean booking additional people.
AI captioning
- Usually priced by usage. SentraCaptions minutes can be bought as you need them or prepaid, with no per-viewer fees and no hardware.
- Can run at every meeting, including committees and commissions that may not have captions today.
- Live translation runs on the same stream: captions and translation in 75 languages, up to 8 per meeting.
- Paused time, such as a recess or closed session, isn't captioned, translated, stored or billed.
For planning help, see budgeting for live captions.
When to use human captioners
Human captioning remains the right choice in some situations, and agencies should plan for it rather than treat it as an afterthought.
- A deaf or hard-of-hearing resident requests CART as an accommodation. Under the ADA's effective communication obligations, the agency should give primary consideration to the aid the person requests. Machine captions are not a substitute for CART in every situation. Take the request seriously and check with your city attorney or county counsel.
- High-stakes hearings with dense technical testimony, where staff want a person making real-time judgment calls.
- Rooms where the audio is poor and can't be fixed before the meeting.
When AI captions make sense
- Every regular meeting, so captions are there by default instead of only on request.
- Committees, commissions and smaller boards that would otherwise go uncaptioned. See captioning commissions and committees.
- Residents who read another language better than English. Translated captions on the chamber screen and on residents' phones support language access in a way a single English caption feed can't.
- Several meetings at once. Multiple rooms can run at the same time, each with its own screen and link.
- Records work. Transcripts in every language are available after the meeting as TXT or SRT files. They are not official minutes, but they can help staff prepare minutes.
Using both
Many agencies settle on a combined approach: automatic captions and translation as the everyday baseline, plus a clear, published process for arranging CART when a resident asks for it. A few practical steps:
- Publish how residents can request accommodations, including CART, and how much notice helps you arrange it.
- Keep automatic captions and translation running even when CART is booked, so the phone page and translated captions stay available.
- Maintain your glossary and audio setup so automatic captions keep getting better.
- Review the ADA Title II web rule timelines with counsel. See ADA Title II and live captions deadlines.
For a deeper comparison, read machine captions vs. CART.