Skip to content
Send a request

PONTAVOX: AI simultaneous interpretation developed in Bern

PONTAVOX development began on 4 August 2026 at the Eventforum Bern, with a prototype built for our own hall. Five days later PONTAVOX ran in front of an audience for the first time; today version 1.4.9 is in use. This page shows how the prototype became an AI interpreter from Switzerland for events with an audience, where PONTAVOX is used and how every new version is tested.

Why an event venue builds its own AI simultaneous interpretation

In the hall, a multilingual event depends on three things. The sound has to come cleanly from the mixing desk, every language needs its own channel to the guests, and someone has to listen from start to finish. The Eventforum Bern has the hall with its mixing desk, a radio system with receivers and a technician on site. Interpreting booths and the equipment have been part of the Eventforum for a long time.

What was missing was an interpretation that fits exactly into this chain. So we built PONTAVOX ourselves. The sound goes from the desk to the laptop running the PONTAVOX Studio, each language leaves through its own output to its own radio transmitter, and guests listen on receivers with headphones, just as with a classic interpreting system. How an event runs is set out under How it works.

The parts we developed ourselves are the technician’s Studio, the path from the mixing desk to the transmitters, the failover watch, the glossary for each event, the client link and the replay. Listening, translating and speaking are handled during the event by a specialised speech service, to which the speech is sent encrypted. Which service it is and where it operates is set out in the data protection sheet under the Swiss FADP, which comes with the quote or earlier on request.

Development from August to November 2026

Just under three months passed between the first prototype on 4 August and the version of 1 November 2026. The timeline shows the main steps.

  1. 4 Aug 2026First prototypeStage sound goes in, one audio channel per language comes out, with a failover watch from the very start.
  2. 9 Aug 2026First use before an audienceAt a church service, PONTAVOX runs live for the first time, in one language.
  3. 30 Aug 2026First event without an alarmFor the first time PONTAVOX runs an entire event on its own, without a single alarm.
  4. Late September 2026New StudioBy early October there is a new Studio for the technician with up to three languages, and the first build that can be installed.
  5. 13 Oct 2026Version 1.2A day after version 1.0 comes the new main chain with a fixed voice, glossary and documents, about four seconds behind the speaker.
  6. 14 Oct 2026First client eventA conference with a panel at the Eventforum Bern runs in three target languages for 2 hours 26 minutes.
  7. 15 Oct 2026The name PONTAVOXUntil then the project ran under the working name KADENZ.
  8. 20 Oct 2026Version 1.3Spanish is added, along with a voice check and a test listen before the start.
  9. 21–22 Oct 2026Security and stabilityWork on security, stability and accessibility follows, plus new voices and read-along for guests.
  10. 23 Oct 2026Version 1.4.0The client link without sign-in arrives, along with a replay that organisers can pass on.
  11. 24–25 Oct 2026Versions 1.4.1 to 1.4.8Deletion after the expiry, sharing by QR code, a guest portal and an interface in four languages are added, followed by a full audit.
  12. 1 Nov 2026Version 1.4.9Volume gets a fixed limit, and the client link can also be shared while the event is running.

In front of an audience since August 2026

Since October 2026, PONTAVOX has interpreted a church service in Bern every Sunday, in English, French, Italian or Spanish depending on the week. There is more on this on the page Churches.

At the Eventforum Bern, the first client event took place on 14 October 2026: a conference with a panel, in three languages at once. So far PONTAVOX has been used in English, French, German, Italian and Spanish. You can choose from 38 target languages; for any that have not yet been used live, we run a trial first.

The Studio shows the distance to the speech for every language. The measured median was 3.96 seconds on 18 October 2026, 4.52 seconds on 23 October and 4.3 seconds over 72 minutes on 1 November. That day, 90 per cent of values were below 4.9 seconds. At a dress rehearsal on 23 October, 9 of 10 technical terms and 3 of 3 names were correct. In a test on 1 November 2026, PONTAVOX also recognised and translated Swiss German well.

To hear what it sounds like, play the samples on the home page.

How a new version reaches the hall

PONTAVOX is being developed continuously; every version has to pass the same tests. What the current version does in the hall is listed under Features. Roll-out follows fixed rules.

  • More than 200 test programsAutomated test programs guard the code, plus 696 individual interface tests, as of 1 November 2026.
  • A test run before every versionBefore a version is rolled out, it goes through a test run.
  • Never during an eventA new version is only installed when no event is running.
  • Backup and a way backEvery roll-out starts with a backup, and the way back to the previous version stays open.
  • Security reviewsVersions 1.3 and 1.4.8 each went through a security review.
  • Less codeWith version 1.4.0, about 100,000 lines of code that were no longer needed were removed. What is not there cannot fail in the hall.
  • Restart with a watchdogIf the Studio crashes, a watchdog restarts it. The interpretation itself never starts on its own.

For venues, AV companies and churches

For now, PONTAVOX is only available for events at the Eventforum Bern, where hall, equipment, technician and PONTAVOX come from one source. How organisers book an event with PONTAVOX is explained under At Eventforum Bern.

If you run a venue, an AV company or a church, you can register your interest for your own hall using the form below. We make no promises. We will get in touch if that changes.

What stays the same: the limits of PONTAVOX

So far, every version has added something. What stays the same is that we state the limits openly and recommend an interpreting team in certain cases.

  • Several voices at onceFor debates with interruptions, and when several people speak at the same time, an interpreting team is the better choice.
  • Negotiations and confidential contentFor negotiations and confidential content we recommend an interpreting team. PONTAVOX is not meant for classified content.
  • Figures and votesThe interpretation is an aid to understanding. The spoken word in the hall is what counts.
  • NamesNames go into the glossary beforehand. We do not promise how they will be pronounced.

For an interpreting team, the Eventforum has two interpreting booths.

Frequently asked questions

Who developed PONTAVOX?

The Eventforum Bern, for its own hall. Development began on 4 August 2026, and five days later PONTAVOX ran in front of an audience for the first time. Version 1.4.9 is in use today, as of 1 November 2026.

Does PONTAVOX have its own AI?

For the interpreting itself, PONTAVOX uses a specialised speech service that listens, translates and speaks; the speech is sent to it encrypted during the event. The Eventforum Bern developed the technician’s Studio, the path from the mixing desk to the transmitters, the failover watch, the glossary, the client link and the replay. A data processing agreement is in place with the speech service, and it does not train on the data.

Was PONTAVOX previously called KADENZ?

Yes. KADENZ was the working name until 15 October 2026. Since then the AI simultaneous interpretation of the Eventforum Bern has been called PONTAVOX.

Where is PONTAVOX already in use?

Every Sunday since October 2026 at a church service in Bern. On 14 October 2026 the first client event took place at the Eventforum Bern, a conference with a panel in three languages. So far PONTAVOX has been used in English, French, German, Italian and Spanish.

How is a new version of PONTAVOX tested?

PONTAVOX has more than 200 automated test programs, and every new version goes through a test run first. It is only rolled out when no event is running, with a backup and a way back to the previous version. PONTAVOX is being developed continuously; every version has to pass the same tests.

Can we use PONTAVOX in our own hall?

Not at the moment. PONTAVOX is only available for events at the Eventforum Bern. If you run a venue, an AV company or a church, register your interest using the form and we will get in touch if that changes.

How far behind the speaker is PONTAVOX?

About five seconds: four to five with fluent speech, five to six with hesitant speech and on panels. On 1 November 2026 the measured median over 72 minutes was 4.3 seconds.

Register your interest

If you run a venue, an AV company or a church, send us a short message. We will get in touch if anything changes. For an event at the Eventforum Bern, choose the topic ‘Event at the Eventforum Bern’.

We only use your details to answer this request.