Hiya Launches Chrome Extension to Detect AI-Generated Voices Amid Election Concerns

A new Chrome extension can reliably detect AI-generated voices

The Hiya Deepfake Voice Detector could come in handy in the final weeks before the US Presidential election. Just in time for the 2024 US elections, the call screening and fraud detection company Hiya has launched a free Chrome extension to spot deepfake voices. The aptly named Hiya Deepfake Voice Detector "listens" to voices played in video or audio streams and assigns an authenticity score, telling you whether it's likely real or fake. Hiya tells Engadget that third-party testers have validated the extension as over 99 percent accurate. The company says that even covers AI-generated voices the detection model hasn't trained on, and the company claims it can spot voices created by new synthesis models as soon as they're launched. We played around with the extension ahead of launch, and it seems to work well. I pulled up a YouTube video about the blues pioneer Howlin' Wolf that I suspected used AI narration, and it assigned it a 1/100 authenticity score, declaring it likely a deepfake. Suspicions confirmed. Hiya threw a well-earned jab at social media companies for making such a tool necessary. "It's clear social media sites have a huge responsibility to alert users when the content they are consuming has a high chance of being an AI deepfake," Hiya President Kush Parikh wrote in a press release. "The onus is currently on the individual to be vigilant to the risks and use tools like our Deepfake Voice Detector to check if they are concerned content is being altered. That's a big ask, so we're pleased to be able to support them with a solution that helps put some of the power back in their hands." The extension only needs to listen to a few seconds of a voice to spit out a result. It works on a credit system to prevent Hiya's servers from getting slammed by excessive requests. You'll get 20 credits daily, which may or may not cover the flood of manipulative AI content you'll come across on social media in the coming weeks.

This Chrome Extension Will Detect AI Deepfakes for You

Hiya can detect deepfaked audio online, including on YouTube, X/Twitter, and Facebook. But it's not always going to be 100% accurate. Have you ever wondered if a video or audio recording was made by AI to trick you into thinking it's real? Well, now there's a Chrome browser extension for that. Hiya is an extension that can detect AI-generated, deepfaked audio just about anywhere online, including on videos or recordings published to X/Twitter, YouTube, and Facebook. It requires a verified email address to create an account in order to access it once installed. Using it involves granting access for each test. The extension only listens to a few seconds of the deepfake before making its decision. It focuses on audio to detect the deepfake, however, so videos without audio won't be able to be checked. Once you've asked it to analyze your active Chrome tab with a video playing, your history of past checked content will appear below in the extension in case you'd like to revisit any results. Hiya's developer says they're releasing the tool now to help prevent political deepfakes from tricking viewers in the weeks leading up to the US federal election. Major political figures have already been extensively deepfaked, from the President Biden robocalls earlier this year to Vice President Kamala Harris deepfakes to AI images imagining Taylor Swift fans supporting Donald Trump. "At Hiya, we're committed to combating scams, fraud, and misinformation," says Hiya President Kush Parikh in a statement. "Deepfakes are becoming harder to detect, making it difficult to discern between real and fabricated content. Voice cloning, while an incredible technological innovation, is also being exploited by cybercriminals for scams and even to try to influence major events like elections." PCMag tested the Hiya extension on a number of videos, including one where deepfakes of Trump and Biden exchange jabs in a sit-down chat. Hiya's extension gave the video a 59/100 on its "Authenticity Score," along with a note: "Our models are uncertain about this voice." That's somewhat reassuring, but it's unfortunate the extension wasn't able to decide with more certainty that the video was a deepfake. It is often accurate, though. A real video of Elon Musk talking about crypto was confirmed correctly as "likely authentic" in a PCMag test using Hiya. The extension was also able to correctly identify obvious deepfakes of Musk as "likely a deepfake" and a deepfake of US Vice President Kamala Harris speaking with her running mate Governor Tim Walz, giving both low trust scores. It identified a video of Will Smith criticizing Biden with 118,000 views as a deepfake, though some users on X seem to have been fooled into thinking it's real. And it identified this video of a supposed former student of Waltz's as a deepfake, as well. If you're unsure about a Hiya result, you can try getting it to analyze multiple different parts of a recording to see if that presents a clearer reading. Otherwise, you'll have to do your own research and generally be wary about sensationalized content published on social media. There are other ways to spot deepfakes, though, like asking a person on a livestream to turn sideways or listening for an off-sounding lilt in someone's voice. Bad deepfakes may be blurry or sound like audio clips being chopped together, but good deepfakes can be harder to spot. In a Hiya survey, 13% of Americans said they've seen a deepfaked video. But more may have seen them without knowing it. In the past, deepfaked YouTube "livestreams" with "Elon Musk" or fake copies of crypto founders like Ethereum founder Vitalik Buterin have duped victims into falling for scams. Access to Hiya is free for now during the election season. It's unclear if or when the company will start charging for access to it at a later date, however, though other deepfake detection tools have already been around for years. YouTube recently announced its own deepfake detection tool for its video site, but it's only going to be available to select creators to start.

New Chrome extension aims to detect deepfake AI-generated voices -- but it's far from perfect

The Hiya Deepfake Voice Detector uses artificial intelligence to determine if the voice on screen is legitimate or faked. The company claims the tool has a 99% accuracy rate and says it can verify (or debunk) audio in just a few seconds. "Our models are trained to detect subtle audio artifacts unique to AI-generated voices -- imperceptible to the human ear but identifiable by machine learning algorithms -- and just a second of audio is enough to detect their presence," Patchen Noelke, vice president of marketing for Hiya, tells Fast Company. I put the free tool to the test and found that while it certainly does catch some deepfakes, its accuracy varied.

Phone spam-fighting startup Hiya rolls out Chrome extension that detects deepfake audio content

Seattle-startup Hiya is joining the growing list of companies using AI against AI to detect misinformation. Hiya announced the launch Thursday of its Deepfake Voice Detector as a free Google Chrome extension that can identify deepfake audio and video content. The company said it needs just a second of audio to determine if a voice is authentic or generated by a voice cloning tool. Deepfakes use voice impersonation and video manipulation to make it appear as if notable personalities -- especially politicians -- are saying whatever the manipulator wants them to say. Advances in technology are making deepfakes harder and harder to detect. Hiya has already seen AI play a key role in its caller ID, call blocking, and spam protection services where generative AI and robocalls are increasing the risk of fraud. The new tool takes deepfake detection wider, analyzing voice content in multiple languages across social media, news sources and other websites. According to the company, Hiya's models are trained to detect subtle audio artifacts unique to AI-generated voices -- imperceptible to the human ear but identifiable by machine learning algorithms -- and just a second of audio is enough to detect their presence. The tool was first revealed in July when Hiya acquired deepfake detection company Loccus.ai. "Deepfakes are becoming harder to detect, making it difficult to discern between real and fabricated content," Hiya President Kush Parikh said in a statement. "Voice cloning, while an incredible technological innovation, is also being exploited by cybercriminals for scams and even to try to influence major events like elections." Hiya said it surveyed 2,000 consumers this year about their exposure to deepfakes, and found that between April and July, one in four people said they'd been exposed to an audio deepfake, with personal voice calls (61%) as the primary risk factor. Social media platforms such as Facebook (22%) and YouTube (17%) were also among the top sources of exposure. Hiya's AI Voice Detection technology is used by Seattle-based nonprofit TrueMedia, which is also working to detect deepfakes and combat disinformation ahead of the election. TrueMedia released its tool to journalists, fact-checkers and others earlier this year, before making the no-cost, web-based tool available to everyone in September. Users can share a social media post containing an image, video or audio file and the AI will analyze the content in real time for evidence of manipulation. True Media and Microsoft are among those who have also created quizzes to test how well people can root out deepfakes. Hiya was a finalist for Innovation of the Year at this year's GeekWire Awards. The startup has raised $42.5 million from investors, with its most recent round coming in March 2023 with a $4.7 million investment, according to PitchBook.