Skip to content
Send a request

PONTAVOX features: AI simultaneous interpretation for events

PONTAVOX provides AI simultaneous interpretation for talks, panels and general meetings in up to four languages, about five seconds behind the speaker, on receivers with headphones. It comes with a glossary for each event, a failover watch, a client link and, if you wish, transcript and audio afterwards. For now, PONTAVOX is only available for events at the Eventforum Bern.

In the hall: AI simultaneous interpretation on receivers

In the hall, PONTAVOX works like an interpreting system. The speech arrives from the mixing desk without room echo, and each language leaves through its own output for its own radio transmitter. The Eventforum Bern provides hall, equipment and technician; the samples show how it sounds.

  • One channel per language

    Choose from 38 target languages, up to four at once; three ran together at a client event. Measured in live use: German, English, French, Italian and Spanish.

  • No app, no Wi-Fi

    Guests select their language’s channel on the receiver. No phone, no account and no interpreting booth are needed.

  • The room stays audible

    When someone on stage speaks a channel’s language, that channel goes quiet and guests hear the room.

  • One fixed voice per language

    The same voice, female or male, from the first sentence to the last, and for everyone on a panel. It is not a copy of a real person’s voice.

  • A muted start

    Whatever PONTAVOX sets by itself starts at half volume at most. By hand, it only goes up to a fixed limit.

Preparation: glossary, documents and trial

It starts with a booking enquiry at the Eventforum Bern: choose the interpreting equipment, switch on the PONTAVOX option, and the quote follows within 12 hours. The glossary guide explains what a good glossary contains.

  • A glossary for each event

    Names, organisations, abbreviations and technical terms, one column per target language, imported from Excel, Word or CSV. At the dress rehearsal on 23 October 2026, 9 of 10 technical terms and 3 of 3 names were right.

  • Documents kept as data

    Programme, speaker list and slide text, up to about 36,000 characters. They are fenced off as data, so no instructions in them take effect.

  • Check and preview

    Before the start, the Studio asks whether everything is right. The technician can briefly listen to each language as a test.

  • Free sound check

    A test tone on each channel checks the sound and the channel assignment. No speech service runs yet, so it costs nothing.

  • Presets for your format

    The Studio has presets for a conference, a general meeting and a panel.

  • Trial in the hall

    By arrangement we test with a page from your programme, and always for languages not yet measured in live use.

During the event: the Studio at the desk

A technician from the Eventforum runs PONTAVOX in the Studio for the whole event and sees every language separately.

The PONTAVOX Studio in operation: three languages with output, level, distance to the speech and volume
The Studio during an event in English, French and Italian. Screenshot of PONTAVOX 1.4.9 with sample data.
  • Each language as a channel strip

    Like a channel on a mixing desk, each strip shows output, level, transcript and distance to the speech. The technician sees at once which language is struggling.

  • The next speaker’s name

    During the event, PONTAVOX is told the name of the next speaker.

  • Mark a moment

    If something sounds wrong, the Studio notes the time and language, so the moment can be found again later.

  • Volume autopilot

    The autopilot suggests levels or adjusts them slowly, by 1 dB per minute at most, and reverses any step that makes things worse.

Failover watch and operational safety

PONTAVOX watches every language separately and acts in stages; the technician is alerted at the first failed attempt. How PONTAVOX has grown since August 2026 is set out under Development.

  • A watch in three stages

    After 6 seconds without data, PONTAVOX rebuilds the language at once; after 12 seconds of silence while others speak, it rebuilds it and catches up on the missed audio. The dead man’s switch at 20 seconds gave 0 false alarms in 6 channel hours of real events.

  • A second path

    If rebuilding fails twice, a second path without the glossary takes over and hands back after three minutes, at the end of a sentence.

  • Handover after 55 minutes

    PONTAVOX renews its connection as a precaution, without losing text.

  • Restart in 1.3 seconds

    By hand, a language restarts in about 1.3 seconds. After a crash, a watchdog restarts the Studio, never the interpretation.

  • Operated only on site

    PONTAVOX can only be operated from the Eventforum’s network. Only the client link and the guests’ read-along are publicly reachable.

  • More than 200 test programs

    Every version must pass the same tests. It is only rolled out when no event is running, with a way back to the previous version.

For organisers: client link, replay and report

You follow every language on your own device. Transcript and audio are available afterwards only with your approval and only from real events, never from trials.

Client view in the browser: choose a language and read the interpretation word by word
The client view via the client link: no sign-in, only for your event.
Share dialogue: link without sign-in, QR code and expiry from 1 to 90 days
Share by link, QR code or email; choose the expiry and withdraw the link at any time.
Replay: choose a language, listen again and read the transcript
The replay after the event: listen and read each language, with downloads.
  • A client link for each event

    It covers only your event, needs no sign-in and is read-only. Share it by link, QR code or email, even while the event is running.

  • Time limit and withdrawal

    Link, transcript and replay stay valid for 1, 7, 30 or 90 days, can be extended and can be withdrawn at any time. Withdrawal and expiry delete the data on the server.

  • A random 192-bit key

    The server only stores a fingerprint of the key in the link.

  • Client view in four languages

    Waiting, live, paused or finished, with every language to read along in the font size you choose. It works in German, English, French and Spanish.

  • Replay

    Listen to each language again with the transcript running alongside and a search, even with the laptop off. Downloads come as MP3, WAV, PDF, Word and text, without the original stage sound.

  • Post-event report

    On request, one A4 page with time on air, availability, interruptions, distance to the speech and glossary, without the transcript.

For guests, and data protection

Guests pick up one of the 120 receivers at the entrance and choose their language’s channel. PONTAVOX processes no data about them; the guide AI interpreting and data protection explains what happens to the speech.

Reading along on a phone: the French interpretation with the current sentence highlighted
Reading along on your own phone, with no app. Guests hear the sound through the receiver.
  • Read along on the phone

    On request, guests read along via link or QR code, with no app or account. Only what the voice has said appears, in step with the voice.

  • No guest data

    No accounts and no tracking; read-along is only counted anonymously. Its text is held only in working memory and, without a release, is gone 45 minutes after the end.

  • Agreement with the speech service

    The speech goes encrypted to a specialised speech service that listens, translates and speaks. A data processing agreement is in place with it.

  • No training

    Data sharing for improvements is switched off; the service does not train on your data.

  • Recording by agreement

    It stays on the laptop at the Eventforum for as long as agreed. On request, there is none.

  • Data sheet under the Swiss FADP

    It names the service, where it operates and what it keeps for how long, with the quote or earlier on request. A sentence for informing speakers and audience comes with it.

Limits: what PONTAVOX deliberately does not do

In these cases an interpreting team is the better choice; it can work in the Eventforum’s two interpreting booths.

  • Interruptions and negotiationsDebates with interruptions, several people speaking at once and negotiations belong in the booth.
  • Confidential contentAn interpreting team is better. PONTAVOX is not meant for classified content.
  • Figures and votesThe interpretation is an aid to understanding; the spoken word in the hall remains authoritative.
  • Pronouncing namesNames go into the glossary, but we make no promise about pronunciation.
  • No sound on the phoneSound comes only through the receiver. We make no promise on how many phones can read along at once.

For now, PONTAVOX is only available for events at the Eventforum Bern. The comparison with the other options is under Cost and comparison.

Frequently asked questions

What can an AI interpreter do at an event?

It interprets speeches in the hall simultaneously into other languages, with a synthetic voice. PONTAVOX puts each language on its own radio channel, about five seconds behind the speaker, with one fixed voice per language. On top come a glossary for each event, a failover watch, a client link and, if you wish, transcript and audio afterwards.

How is PONTAVOX different from a translation app?

With a translation app, guests listen or read on their own phone. With PONTAVOX they listen on receivers with headphones, and a technician looks after every language at the desk. Glossary, failover watch, client link and replay are part of it. The Cost and comparison page also covers remote interpreting platforms and interpreting teams.

Can the organiser follow the interpretation live?

Yes, with the client link, on your own device and without signing in. The link covers only your event and shows the transcript of every language in a font size of your choice. You set the time limit and can withdraw the link at any time.

Do we get a transcript after the event?

Yes, transcript and audio of each interpretation, if you release them. In the replay you listen to and read every language again and download it as MP3, WAV, PDF, Word or text. The original sound from the stage is not included, and the server deletes the replay automatically when the time limit expires.

What happens if a language fails?

PONTAVOX rebuilds it as soon as 6 seconds pass without data, or when it stays silent for 12 seconds while others speak, and catches up on the missed audio. If that fails twice, a second path takes over. The technician is alerted at the first failed attempt.

How long is the delay with AI interpreting?

About five seconds: four to five with fluent speech, five to six with hesitant speech and on panels. The measured median was 3.96 s on 18 October, 4.52 s on 23 October and 4.3 s on 1 November 2026. That day, 90 per cent of values over 72 minutes stayed below 4.9 s.

How does an AI interpreter learn the names and terms of our event?

Through the event glossary, with one column per target language, imported from Excel, Word or CSV. The programme, speaker list and slide text can be added, up to about 36,000 characters. At the dress rehearsal on 23 October 2026, 9 of 10 technical terms and 3 of 3 names were correct.

Where can PONTAVOX be booked?

For now, only for events at the Eventforum Bern. In the booking enquiry you choose the interpreting equipment and switch on the PONTAVOX option. The quote follows within 12 hours.

Plan your event with PONTAVOX

Tell us the date, the languages and the type of event. The Eventforum Bern will send your quote within 12 hours, with the data protection sheet if you wish.

We only use your details to answer this request.