How does an AI interpreter work?
An AI interpreter runs three steps, over and over, without a break. It hears what is said on stage, translates the content into the target language and speaks the translation in a synthetic voice. All of this happens while the speaker keeps talking, as with simultaneous interpretation in a booth.
How good AI interpretation sounds depends first on the audio. Room echo, applause and background noise make speech recognition harder. That is why PONTAVOX takes the sound straight from the mixing desk, clean and without room echo. It starts before the end of a sentence and waits when the speaker hesitates.
- ListenThe mixing desk passes the microphone sound to the laptop running the PONTAVOX studio.
- TranslateA specialised speech service translates the speech continuously into each chosen language.
- SpeakEach language has one fixed voice, female or male, from the first sentence to the last.
How PONTAVOX differs from a phone app
Many AI interpretation tools for events run on the guests’ own phones. Guests scan a QR code, open a page and plug in their own earphones. “No app” then often means a link in the browser. Every guest still needs a charged phone, earphones and a network connection in the hall.
PONTAVOX takes the path of a classic interpreting system. Each language leaves the laptop as its own line-level signal and goes to its own channel. That can be a radio transmitter, an infrared radiator, an amplifier with loudspeakers for a side room in one language, an interpreter console or the audio of a livestream. The page Equipment and connection shows how it connects.
- No phone neededGuests choose their language channel on the receiver and listen on headphones. No app, no Wi-Fi and no interpreting booth.
- The room stays audibleWhen someone on stage speaks a channel’s language, that channel goes quiet and guests hear the room.
- A technician at the deskIn the PONTAVOX studio they see every language with its output, level and distance to the speech.
- FailoverIf a language goes quiet while someone is speaking on stage, PONTAVOX rebuilds it and catches up on what was missed. If that fails, a second path per language takes over.
- TestedMore than 250 automated checks run before every new version.
How long is the delay in AI simultaneous interpretation?
With PONTAVOX the interpretation runs about five seconds behind the speech: four to five seconds with fluent speech, five to six with hesitant speech and on panels. In the audio samples on the home page you can hear when the interpretation comes in and how it keeps pace.
A panel works if everyone has their own microphone and speaks one at a time. When several people talk at once, human interpreters are the better choice.
On stage, PONTAVOX needs standard German or English. There are 38 target languages to choose from. As many run at the same time as the system has channels, with one output and one channel per language. German, French, English and Italian have been measured on real speeches; for other languages we run a trial first. The full list is under Languages.
Why an AI interpreter needs a glossary
An AI interpreter does not know your speakers, nor the abbreviations everyone in your association takes for granted. For technical terms it needs to know which translation your organisation uses. So we prepare every event. The programme, speaker list, names, abbreviations and technical terms go into the glossary beforehand, so PONTAVOX does not guess names.
You can also send documents such as slides or a manuscript in advance. The more we know before the event, the more complete the glossary.
When an AI interpreter fits, and when human interpreters are better
An AI interpreter fits when people on stage speak one after another. That covers talks, presentations and welcome addresses, as well as panels where everyone has their own microphone. The speech is in standard German or English.
Human interpreters are the better choice for Swiss German on stage, debates with interruptions, negotiations and confidential content. We would rather tell you before the event than after it.
How a team works in the booth is explained on Simultaneous interpretation explained. What the different options cost is set out under Cost and comparison.
| PONTAVOX | Interpreter team | |
|---|---|---|
| Talk in standard German or English | fits | fits |
| Panel, one microphone per person | fits | fits |
| Swiss German on stage | not possible | fits |
| Debate with interruptions | team is better | fits |
| Negotiation, confidential content | team is better | fits |
| Booth | none | one per language |
| Voice | one fixed voice per language | changes with the interpreter |
PONTAVOX is not meant for classified content.
Data protection: what happens to the speech
During the event the speech is sent encrypted to a specialised speech service that listens, translates and speaks. Which service it is, where it operates and what it keeps for how long is set out in our data protection sheet. You can have it on request before the quote.
The contract is made in Switzerland, in Swiss francs. For confidential content we recommend human interpreters.
Where you can use PONTAVOX
PONTAVOX runs at real events at the Eventforum Bern. There you book hall, equipment, technician and interpretation from one source, with a quote within 12 hours. In Bern, PONTAVOX is only available at the Eventforum Bern.
Elsewhere we license PONTAVOX for the duration of your event. The venue, the AV company or the organiser runs it on site with their own equipment. We do not travel with equipment and technicians. We agree terms, set-up and support with you in a conversation; the easiest way to start is the form below.
After the event you can have a customer sheet with the figures for each language. The page How it works goes through an event step by step.