AI translator can make you seem instantly multilingual
This week, Google unveiled PaLM 2, an AI system that can easily translate languages, enabling users to access more of the internet. But a Canadian company has found an even more impressive use for multilingual AI: translating video calls.
Fibernetics, a telecom provider based in Cambridge, Ont., has developed software that enables almost real time voice-to-voice translation of video calls. Using the technology, an English-speaker connecting to a francophone meeting would hear an English translation of other participants. And, when the English-speaker opens their mouth, the others would hear fluent French.
The AI translation will be part of a new business communications platform, called Nucleus, that Fibernetics launched this week (the translation system will be added in the summer). It will also feature other AI-powered innovations, such as an “eye focus” system that
automatically adjusts a user’s image so it appears as if they are looking directly into the camera, even if they’re glancing at notes on a second screen.
Fibernetics co-founder John Stix says 4,000 small businesses, entrepreneurs and startups have already signed up to the service. We spoke to Stix about this new technology — and how it could change business meetings forever:
What kind of difference will realtime translation make?
Cultural and language barriers can be a real challenge for businesses. Think of a small accounting firm in Quebec. Because of the language barrier, they could only operate regionally.
Now, through Nucleus, they can jump on a video call, select the language they want and have a conversation with English-speaking clients anywhere in the world.
It levels the playing field a bit.
Exactly. There are almost 100,000 new small businesses launching in Canada every year. We want to help them reach beyond our borders and access the international talent pool. Let’s enable a startup here to have a talk with software developers in Peru or Nicaragua.
How many languages are available?
Just French and Spanish to start. We hope to have Hindi in the next six months. Eventually, Nucleus will be able to translate 10 major languages. Our dream, on a whiteboard somewhere, is to one day include sign language.
Does it translate in the speaker’s voice?
For now, you’ll hear an AI-generated voice. But we’re not far off from solving that. In the next six months to a year, I should be able to give a 30-minute voice sample and it will learn what I sound like.
How accurate will the translation be? AI transcription can be a little wonky when it comes to things like dialects and accents.
The more I speak, the more the AI will learn how I speak. Eventually, it will pick up on most of the subtle nuances of my speech. There are many tiny important ways we communicate and AI will eventually adapt to them.
And this goes beyond speech, doesn’t it?
More than 70 per cent of our communication is non-verbal. This is why Nucleus’s eye-focus capability is so important. We need to make eye contact with each other. When we talk, we consciously or unconsciously pick up on cues like eyebrows going up and down. Eye-focus gives you the freedom to read from a separate screen and not lose contact with your audience. It looks totally natural. It even mimics blinking.
Where else could this technology be used?
There’s no reason we can’t one day launch an app for households. My wife is French Canadian and she has a lovely grandmother who I’ve never been able to say more than a few words to. I should’ve learned French a long time ago, but didn’t. It would be fantastic to have a real conversation with her.