Once purely in the realm of science fiction, the idea of a universal translator has gradually become reality thanks to advancements in artificial intelligence.
Tech giants like Apple, Microsoft, and Google have all developed technologies to translate text or speech in real-time in a browser, on a mobile app, or even on video call software like Skype.
For the past few months Australian telco Optus has been plugging into Google’s translate API to beta test live, automated translations of calls made over its mobile network.
Around 1,200 people have joined the free beta with translated phone calls ranging from a few minutes to over an hour.
“We’re seeing Italian, Mandarin, Vietnamese, and Hindi being some of the most popular languages, unsurprisingly, and we’re getting phenomenal feedback from customers,” Optus VP of Product Development, Clive Dickens, told Information Age.
“This is a really important problem it solves for millions of Australians who don’t have English as their first language and if they want to make an appointment for a vaccine, or they want to order takeaway, or they just want to arrange for a tradie to fix something.”
It works by first transcribing the call in its original languages then sending the text to a Google Cloud API which translates the text before it is sent back out as audio.
Calls tend to take longer, and it takes some adjustment for people on both end as they get used to a different mode of speaking, but the result could be a whole new way of interacting with businesses and clients in Australia and around the world.
Tech giants have struggled with balancing quality assurance and privacy with voice-activated products – after all, if you want to know that your AI transcriptions or translations are working properly, the easiest way is to hire people to do just that.
Microsoft came under fire in 2019 after it was discovered the company paid contractors to listen to snippets of personal conversations made through Skype’s translation features in order to improve the product.
Optus has said it doesn’t have direct access to the transcripts and is relying on user feedback and its pool of employees testing the product to know how well it works with Dickens saying Google translate was “as good as it gets”.
Likewise, Google said there’s no risk of people reading private conversations from Optus’ plugin with its API.
Matt Zwolenski, the Director for Customer Engineering at Google Cloud Australia, told Information Age it was a simple workflow that doesn’t involve any extra data siphoning.
“We get fed the piece [from Optus], keep it for a very short period of time – purely for that service through the API – and then we delete it,” he said.
“There’s obviously a short period where it’s held for if there are issues reported or debugging needed, but fundamentally the function we provide is: receive something, translate it, send it back, delete.
“Everything on top of that is really managed by Optus so there’s no risk of us listening into calls or for anything like that to occur.”
Exactly how much Optus is paying for that function remains undisclosed, but Google does publish pricing information for its Cloud Translation service.
Access to the API is free for the first 500,000 translated characters each month. After that, Google charges US$20 for every million characters it processes.
Optus has not yet decided how it will turn Call Translate into a paid-for service once it goes into production and eaves the beta period which is expected to happen early next year.
“We really want to know how our customers value the service before we put a cost on it,” Dickens said.
“We can really see great examples of small and medium businesses wanting to use this, whether it’s tele sales, care, appointment settings, generating leads, or inbound inquiries.
“This is about the mobility of voice, particularly when you don’t necessarily know the person on the other side, but you do know that you don’t speak the same language.”
