Your interpreters and AI, on the same broadcast
Give one language to a person and the rest to AI. Listeners pick whichever they prefer, in the same session, and for as long as they stay with the person, no model runs on that language at all.
At a glance
- AI cost on an interpreted language
- Zero
- Interpreter equipment
- Browser and headset
- They can work from
- Anywhere
- Organization data they can see
- None
The interpreter is a first-class part of the broadcast
An interpreter opens their own page, takes one output language, and speaks. Their voice is published into its own room for that language and no model sits in that path at any point β the audio goes from their microphone to the listeners. Everyone on that language hears the person by default, and can switch to the AI with one tap.
- One interpreter per language, held by their connection
- Listeners switch between the person and the AI
- If the interpreter drops, listeners move to AI on their next reconnect
- No AI seconds while listeners stay with the person
Good morning everyone, and thank you for joining today's global session. It's wonderful to see so many of you connecting from around the world.
Why put a person on a language
Not because the AI cannot do it β because sometimes a person should.
The bill drops where you have people
No AI session runs for that language while listeners stay with the person, so it is translated for free. A church with two volunteers pays for the languages they cannot cover, and little more.
Volunteers are first-class
An admin invites an interpreter by email address. They claim the invitation the first time they sign in, with no account set up in advance.
A person on the language that carries risk
Where a mistranslation would be a liability, put a professional there and leave the AI to carry the rest of the room.
They can monitor the room from home
A remote interpreter can ask for the source audio back through their own connection, so they do not need to be in the building.
Their own record of what they did
Every stint on air is recorded for the interpreter, summed per service β without exposing the organization's costs or audience figures to them.
Hand over without stopping
An interpreter can take a language mid-service, release it, and let the AI pick it back up. The broadcast never pauses for the swap.
Setting up an interpreter
- 1
Say which languages a person may take
In the organization settings, mark the output languages that are open to interpreters. It is a subset of the languages the broadcast already carries.
- 2
Invite them by email
The invitation is addressed to an email address and claimed by whoever confirms that address. An interpreter seat opens the interpreter screens and nothing else.
- 3
They take a language and start
One interpreter holds a language at a time. Reloading their own page gives it straight back to them; a dropped connection releases it.
- 4
Listeners choose
On an interpreted language the person is the default. A listener who prefers the AI switches, and anyone left on the person moves to AI automatically if the interpreter stops.
Common questions
Can two interpreters share one language?
Not at the same time. One interpreter holds a language while their connection is open. When they stop or disconnect, the language is free for the next person β or falls back to the AI.
What happens if the interpreter has to stop suddenly?
Their channel closes, which makes the listeners' players reconnect, and on that reconnect they are given the AI translation of the same language. Nobody has to do anything.
Can an interpreter see our sessions, costs or audience?
No. An interpreter account reaches the interpreter screens and its own history of stints on air, and nothing else. Sessions, credits, transcripts, recordings and audience figures are not readable from that account.
Does an interpreter need special equipment?
A browser and a headset. No booth, no console, no software to install. They can be at the back of the room or in another country.
Does using an interpreter really reduce the bill?
Yes, with one thing worth knowing. Translation is charged per minute per language, and while your audience stays with the interpreter no AI session runs on that language, so it accrues no charge. A listener who switches to the AI opens one, and that language then bills normally for as long as somebody is there β so the honest figure is "nothing while they stay with the person", not "nothing, full stop".
Can we start with AI and add people later?
That is the usual path. Run everything on AI first, see from the audience figures which languages actually matter, and put your volunteers on those.
Have interpreters already?
Bring them onto the same broadcast and let AI cover the languages they cannot.