AI translation & radio

A shared language for joint operations

One officer speaks Chinese; the liaison reads English captions and hears English voice within two seconds — and the reverse. The same engine turns an off-the-shelf handheld radio into an on-premises interpreter. No audio ever leaves your room.

How it works

Three AI engines, one server, zero cloud

1

Listen

On-premises speech recognition for Chinese and English, tuned for narrow-band radio audio.

2

Translate

An on-premises translation model, with terminology lists you control for military, financial or semiconductor vocabulary.

3

Speak

Tessavox Voice, our own engine, speaks the translation — only to the participants who chose that language.

4

Each to their own

Every participant picks a caption language. While translated voice plays, the original audio is automatically lowered.

A responder speaking into a handheld radio
Radio-to-radio

An interpreter on the channel

In combined exercises, language is the obstacle on the radio net. Connect one off-the-shelf handheld radio through a USB interface and Tessavox translates every transmission on the same channel.

  • Follows radio procedure: “完畢 / Over” hands the channel over
  • Transmits only when the channel is clear, to avoid doubling
  • Transmit safety: time-limited keying, automatic release on any fault
  • No radio modification; works in places with no network at all
Tessavox Voice

Our own voice engine

Translated speech is spoken by Tessavox Voice, a speech synthesis engine we trained ourselves. Trained on Taiwanese Mandarin, it reads Chinese, English and mixed Chinese–English sentences in one pass. It runs on the server's CPU — no GPU and no network required.

  • Taiwanese Mandarin: polyphonic characters, tone sandhi, numbers and units read the way Taiwan reads them
  • Mixed Chinese–English sentences in one pass; pronunciations of terms and acronyms can be customized
  • A dedicated voice: built from recordings of a person you designate, with the built-in recording tool (about 1–2 hours of speech)
  • Models and training data under our control — no speech models or components developed in mainland China

Internal test, October 2026: 1.5–1.9 s from the end of a sentence to translated voice playback (3 Chinese and 3 English sentences); a dedicated Chinese voice built from about 100 minutes of recordings. Dedicated voices are built with our assistance.

Seminars & briefings

Live captions on every attendee’s phone

Delegation briefings, vendor tech sessions, international seminars: the speaker talks in English and the audience reads Chinese on their own phones — or the other way round. Scan the QR code at the venue: no app, no login. Speech and speaker materials stay on your server.

  • No app to install, no login
  • Chinese or English, with the original alongside and adjustable text size
  • Consistent terminology from the glossary you approve before the event
  • The event code works only during the event; attendees can only receive
  • Bilingual transcript download afterwards
1

Prepare

Upload the speaker’s slides, script and reference documents. Tessavox suggests terms with Taiwanese translations for you to approve, and translation also consults the materials.

2

Capture

Connect the podium computer to the mixer or a microphone and open the capture page — with a level meter and a live transcript.

3

Scan and read

Attendees join the venue Wi-Fi, scan the projected QR code and pick Chinese or English. Late arrivals see the most recent captions right away.

Internal test, October 2026: a public talk recording as the speaker and an Android phone as the attendee — the Chinese translation appeared about 0.5 s after each passage. Requires a server with a GPU (Standard or High-performance tier).

Everything joins the picture

Meetings, radios and TAK chat — translated together

Radio in the meeting

Radio traffic appears in the command meeting as original audio plus bilingual captions. English instructions from the room are translated and transmitted in Chinese.

TAK chat, both ways

What is said in the meeting reaches every ATAK as bilingual text; English typed in the field becomes Chinese captions and voice in the room.

Terminology you control

“second element” → 第二梯隊, “QRF” → 快反部隊. Apply lists to specific rooms, radios and TAK messages; changes take effect immediately.

Governed like the room

Captions and translated voice follow the same room access rules. No permission to the room, no captions either.

Transcripts for recordings

Recorded calls and rooms are transcribed on the server, with speakers labelled and searchable text. More →

Desktop and browser

Captions and translated voice work in the browser and in the Windows and macOS desktop apps.

Why on-premises

Cloud meeting tools translate too. Here is the difference.

Typical cloud meeting translationTessavox
Where speech is processedThe vendor's cloudYour own server room
Disconnected or air-gapped networksNot availableWorks fully offline
Handheld radios, SIP and H.323—Supported
Military terminology, radio procedureGeneral business useCustom terminology; “Over” hand-off
Translated voiceThe vendor's stock voicesOur own Tessavox Voice; dedicated voices available
PricingPer-user subscriptionPer project

Cloud products offer more languages and suit general business use. Tessavox is for organizations where speech must not leave the building and radios and legacy systems must be included.

See it live — on our hardware or yours

A 60–90 minute demo: command-center layout, a drone on the TAK map, radio translation and the phone system, shown end to end. Online, or at your site fully offline.