Google’s Pixel 10 smartphone can deepfake your voice on-device

As recruiters, financial institutions and police battle a wave of scams and cybercrime fueled by the imitative power of AI-enabled deepfakes, some consumer products are actually leveraging it as a unique selling point.

Google has been touting its latest smartphone, the Pixel 10, which comes with a host of upgrades. Tech reviewers have been quick to note that many new features come via software. One headline-grabbing upgrade is a native voice deepfake capability for the Pixel 10’s new real-time translation feature.

While universal translators have thus far been the preserve of sci-fi shows, translation apps in the real world have come a long way. However, they still don’t work with the seamless, instantaneous ability that would allow two people of different languages to have a real-time conversation with each other.

At its Made by Google keynote, Google showed off something very close. The new feature not only translates speech in real time but also deepfakes the person’s voice, in near real time, so that the other person can hear your voice speaking in their native language. Star Trek, eat your heart out! And it works for both speakers, so that there can be two deepfaked voices speaking to each other.

Gizmodo’s Senior Editor for consumer technology, Raymond Wong, captured a live demo of the feature at Google’s event, showing American talk show host Jimmy Fallon having his voice translated, deepfaked, into Spanish via the Pixel 10. Wong posted the video on his X account, and Google’s deepfakery appears to work well, not only nailing the translation but capturing Fallon’s voice and its cadences.

The feature is powered by Gemini Nano, a leaner version of Google’s large language model (LLM), and the Pixel 10’s Tensor G5 chip. Google says the Nano and the translation, deepfakery and all, are processed on-device, meaning your voice, and whatever it says, does not get beamed into the cloud.

This should allay privacy fears (at least a bit) since having biometric information like your voice stored on the cloud somewhere could potentially be very risky considering people’s voices are used for security purposes in various contexts, such as banking.

Deepfaked voices are a growing problem in the business world where corporations have been tricked into making fraudulent transactions. If you hear your boss telling you to proceed in processing an invoice, who would say no when the voice sounds so genuine? AI is enabling such illegal activities and Google’s version, Gemini, is a clear indication of how far the technology has come in its impersonation capabilities.

Article Topics

deepfakes  |  Google  |  Pixel 10  |  smartphones  |  synthetic voice