Simultaneous interpretation without booths or receivers
One machine at the production desk listens to room audio and translates it while it is being said. It leaves by two doors at once: inside your video signal, and on every attendee’s phone.
- languages in the catalogue
- 58languages in the catalogue
- at once in one event
- 12at once in one event
- receivers to hand out
- 0receivers to hand out
- with a real alpha channel
- NDIwith a real alpha channel
What is AI simultaneous interpretation?
Interpreting a speech while it is being delivered rather than waiting for the speaker to finish. The classic version is done by human interpreters in a soundproof booth and heard through radio receivers. The AI version does the same job with speech recognition and machine translation: the audience reads (or listens to) the result on their own phone, and there is no hardware to hand out or collect.
The problem today
- A booth is quoted per language per day, with receivers per unit on top and somebody counting them every evening.
- Each added language is two interpreters, because they need relief. The cost does not rise gently. It steps.
- Breakout rooms get no interpretation at all, because the budget went to the plenary.
- Anyone hard of hearing gets nothing: interpretation solves language, not access.
How it joins your production
Room audio comes in on the card
The same feed you already send to record. One, two or four microphones by plan.
They leave as NDI with a real alpha channel
Your switcher composites them over programme as one more input.
Or full-screen HDMI with a chroma key
For when the venue screen is not behind a switcher.
The audience joins by scanning a QR code
No app, no account, no sign-up. Any browser on any phone.
Each attendee picks their own language
Up to 12 at once, from a catalogue of 58. The person reading decides.
What the attendee gets
The text in their language, on their own phone
And they can change language at any point during the event, as often as they like.
Audio, if they prefer it
The phone reads the translation aloud through their own earphones.
Questions and votes
They ask from the phone; whoever moderates decides which questions go up on the venue screen.
No per-attendee or per-language charge
Five hundred people in twelve languages costs what twenty in one costs.
Four questions worth asking
Does it go inside your video signal?
Here it does: an NDI output with a real alpha channel in the frame, composited over programme.
What happens when the connection drops?
The audience output fades to transparent. Never a stale caption or an error card.
Does it charge per attendee or per language?
Ask it before the hourly rate. It decides whether the third language gets approved.
Where is it processed and what is stored?
Audio is transcribed in your account’s processing region (Europe, the Americas or Asia-Pacific) and translated by OpenAI in the USA. We store no transcript on the live path: only if you ask for the AI minutes or keep the transcript when closing the app, and it is deleted after 30 days. The sub-processor list is published.
Event Pro
Translation into all 58 catalogue languages and two microphones, which is the ordinary shape of a panel or an interview. Broadcast Multi-Mic when four people speak at once.
Frequently asked questions
Does it replace a human interpreter?
The honest claim is that it replaces the interpretation nobody budgets for today: the second, third and fourth language, which almost no event books at all. Where interpretation IS booked there is usually a reason (a legal or diplomatic negotiation where nuance carries consequences) and there you book a person.
How many people can follow at once?
Measured against the deployed system: five thousand concurrent phones in one room with no refused connections. We recommend selling five hundred to a thousand per event, because above that latency starts approaching our own quality target. The real limit at a conference is usually the venue Wi-Fi, not us.
How long until the translation appears?
Speech reaches our service as text in 718 ms median and 1,038 ms at the 95th percentile, measured against the deployed service; untranslated captions are written on screen shortly after. Translated captions land about 1.7 s after the word (median), and 2.5 s in the slowest one in ten.
Do we have to hand out receivers?
No. Every attendee uses their own phone and their own browser. There is nothing to issue, count, charge or clean.
What about a closed-door session?
The room code is not a confidentiality control: anyone holding it can read along. For a private session, use the venue screen only and do not circulate the code.
For your kind of event
Other ways people search for this
Try one hour free
One hour of captioning and translation on one microphone, no expiry, no card. Enough to prove it on your own rig before you pay.