An AI voice interprets the speaker live: it hears what is said and speaks it in the listener's language about 1.5 seconds later. It is not the captions read aloud.
AI voices · AvailableChoose the AI voice
In the session form, under Interpretation, the AI voice settings come in one of two forms, depending on your account.
If your form shows an AI voice choice with hours
You choose one AI voice option:
- No AI voice: captions and text translation only, in every language. Uses no hours.
- Languages you choose: only the languages you pick get a voice. Best when you know your audience. Add them with Add a language.
- Any language: listeners pick from all the AI voice languages on their phone. A voice starts when someone picks it.
AI voices use voice hours, and only while someone listens. The form shows how many hours the session is likely to use and whether your balance covers it (Covered or May run short).
If you're out of AI voice hours, the session has captions only. Captions and your interpreters still work. Select Add hours, or ask your account owner to add them.
If your form shows AI voice languages
Under AI voice languages, select Add a language for each language the AI should interpret the speaker into. Only the languages named here get a voice; every other language has captions. Your plan sets how many languages a session can have, and the form shows how many remain. If AI voices are not in your plan, the form shows Not in your plan.
When a voice starts
A voice starts when a listener picks its language. InterScribe can enable starting voices at go-live for your account on request; your form then shows Start the … voice at go-live next to each voice.
Make the voice sound like the speaker
In some languages the AI voice can sound like the person on stage. Turn on Make the AI voice sound like the speaker under the voices. Other languages keep their usual AI voice.
Before it turns on, you confirm that you have permission from every person who will speak at the event to clone their voice. Know this first:
- The copy is built live from their speech during the event. InterScribe does not keep a voice model of any speaker.
- The speakers' audio is processed by our AI provider, Alibaba Cloud, in Singapore, under its data-protection terms.
- The cloned voice takes about 3–4 seconds to start speaking, instead of about 1.5 seconds with a standard voice.
- If the voice provider is not available at the moment, listeners may hear a standard voice instead.
- Listeners see a small icon next to the language that tells them the voice is a synthetic copy of the speaker's voice.
- If the event is recorded, the interpretation is recorded in the cloned voice.
You can turn it off at any time. Languages added after you turned it on keep their usual AI voice until you confirm permission for them.