How speech-to-text, machine translation, and text-to-speech route back out to Unite receivers over Dante.
By Mark Hall, National Product Marketing Manager, Corsair Solutions
Televic Unite AI translation drew a steady stream of questions on the stand at Integrate, usually with the same one behind it: where does the translated audio actually go? Live captions on a phone screen are one thing. Getting synthesised speech back into the receivers people are already wearing is another.
Here is how the workflow fits together.
The Unite platform
Televic Unite runs on transmitters and receivers. There is a master, a submaster, and a range of receiver units. They work standalone, or through an access point when you need greater distance or multiple groups running at once.
The flexibility comes from the connectivity. Unite is Bluetooth in and out, and USB in and out. In a classroom the teacher might hold the master at the front, with a submaster at the back of the room, and students listening on their own Bluetooth headphones. Or the receiver connects by USB-C into a laptop and feeds straight into Teams or Zoom.
Getting the audio in
The presenter speaks into the master transmitter, the same device used for any Unite session. That transmitter connects by USB to an Intel NUC, where it appears as a Unite microphone device. The NUC streams to the Wordly cloud platform, which runs live speech to text and translates into the selected target languages. Dutch and Simplified Chinese were running in our setup.
Input quality is the thing that makes it work. Unite delivers clean, per speaker audio rather than a room mic collecting everything, and transcription accuracy depends heavily on what it is fed.
Getting the audio back out
This is the part that makes it speech-to-speech rather than a captioning service, and it is what turns the workflow into a working interpretation system.
- Wordly generates a live audio stream for each language using text to speech
- The NUC runs Dante Virtual Soundcard, which puts those translated streams onto the Dante network
- DVS routes each language over Dante to the Unite access point, which supports Dante natively
- From there it works as a normal Unite system, with each language on its own channel
Multiple languages run simultaneously. Delegates select theirs on the receiver, the same as they would with interpreters in booths.
Dante is not the only way to do it. If there is no Dante network available, you can take the output from the computer directly into a transmitter on a separate channel instead. Fewer moving parts, though you give up the channel count and routing flexibility that Dante gives you once you are running several languages at once.
Where it fits
This does not replace human interpretation everywhere. Formal proceedings, legal settings, and anything where nuance carries consequences still warrant an interpreter. Where it earns its place is the middle ground: council meetings, corporate sessions, training rooms, and community events where interpretation was previously out of reach on cost or availability.
The practical constraints are worth naming too. There is latency through the chain, the cloud leg needs a reliable connection, and output quality tracks input quality closely.
Seeing it in action
We demonstrated this setup at InfoComm Asia in Bangkok, then ran it live on the stand at Integrate in Sydney. Wordly is a separate cloud platform that Unite feeds into.
If you have multilingual requirements on a project, get in touch and we can talk through whether this approach suits.
Watch Mark Hall walk through the Unite setup on the Televic Conference pod below.