{"id":84845,"date":"2025-08-21T11:39:09","date_gmt":"2025-08-21T11:39:09","guid":{"rendered":"https:\/\/www.newsbeep.com\/au\/84845\/"},"modified":"2025-08-21T11:39:09","modified_gmt":"2025-08-21T11:39:09","slug":"googles-big-leap-forward-for-real-time-translations-is-deepfaking-your-voice","status":"publish","type":"post","link":"https:\/\/www.newsbeep.com\/au\/84845\/","title":{"rendered":"Google’s Big Leap Forward for Real-Time Translations Is Deepfaking Your Voice"},"content":{"rendered":"

I can\u2019t count the number of times I\u2019ve been promised the Next Big Thing in app-enabled translation. Since the dawn of Google Translate<\/a>, it\u2019s felt like tech companies (not just Google, but Samsung<\/a> and Apple<\/a>, too) have been teasing a future where speech can be translated instantaneously, allowing for near-seamless communication between people from *sarcastic SpongeBob voice* across the world. But the truth is, that future, however enticing on paper, hasn\u2019t exactly panned out.<\/p>\n

As incredible as apps have gotten at translating speech and text with a fairly high degree of accuracy, they haven\u2019t quite risen to the speed and cadence of real-life conversations. Designing a translation tool that can keep pace with our mouths (like, actually talking) isn\u2019t an easy feat. We talk fast, and we expect even faster responses, which makes live translation less of a marathon than a sprint, or I guess more accurately, a sprint that could be a marathon in length.<\/p>\n

See Pixel 10 Series at Amazon<\/a><\/p>\n

See Pixel 10 Series at Best Buy<\/a><\/p>\n

Given that long promise of snappy, useful, real-time translation, I\u2019m conditioned to roll an eye or two when live translate enters into a keynote, which is exactly what I did during Google\u2019s annual Pixel hardware event. This year, though, that eye roll might not be warranted. At its Made by Google keynote,<\/a> Google showcased a feature that not only translates your speech in real time, but also deepfakes your actual voice (also in near real time) so that the person on the other end can hear you speaking in their native language. And yes, it works in the inverse, too. That\u2019s right, just two deepfakes talking to each other; nothing to see here, folks.<\/p>\n

And the extra wild part is Google was so confident in its new live translation feature that it offered up a live demo, which, I\u2019m not going to lie\u2026 it kind of nailed? Gizmodo\u2019s Senior Editor, Consumer Tech, Raymond Wong, captured the whole thing live at Google\u2019s keynote. For your viewing pleasure, Jimmy Fallon\u2019s voice deepfaked into Spanish:<\/p>\n

Wow, Jimmy Fallon’s voice was translated\u2014deepfaked\u2014into Spanish on Pixel 10\u2026 and it worked fast and accurately. Even getting inflections correct. This is the quiet game-changing AI feature\u2026 translations pic.twitter.com\/SAXeowCXpE<\/a><\/p>\n

\u2014 Ray Wong (@raywongy) August 20, 2025<\/a><\/p>\n

I was also watching along from home during this segment, and my partner, who\u2019s Spanish-speaking and bilingual, confirmed that Google\u2019s new AI translate feature seemed to ace the assignment, inflections and all. Don\u2019t get me wrong, I still want to test those translation features for myself, but from the looks of it, Google is off to a pretty amazing start here.<\/p>\n

Powering those translation abilities is Gemini Nano<\/a>, a compact version of Google\u2019s increasingly iterated-upon large language model and the Pixel 10\u2019s<\/a> Tensor G5 chip. Google says its Nano model and the translation feature are run on-device in this case, which means that nothing\u2014including your calls\u2014gets teleported to the cloud while you\u2019re translating. That makes its new feature less icky, and I say \u201cless\u201d in this case because let\u2019s be honest, this thing is still deepfaking your voice.<\/p>\n

Truthfully, if Google\u2019s new translation feature wasn\u2019t happening on-device, I might be a little worried. As cool as it is, the thought of having a facsimile of your voice stored on a server somewhere is a bad one, given the fact that people use biometrics for all sorts of important digital security, banking included. And in a way, on-device or not, the feature is still creepy. Apparently we\u2019re at the stage of instantaneous voice deepfakes. Just imagine what AI can do with a little bit of time and training.<\/p>\n

But more than anything, I\u2019m impressed by what Google showed off today, especially as someone who\u2019s watched tech companies overpromise on translation features for years at this point. It\u2019s still too early to declare that Google has hit the Holy Grail of real-time translation, but for once, I\u2019m left thinking that the idea of seamless, phone-enabled translations has actually taken a major leap forward. So, consider my eye roll officially rescinded, Google.<\/p>\n

See Pixel 10 Series at Amazon<\/a><\/p>\n

See Pixel 10 Series at Best Buy<\/a><\/p>\n