A 15-second sample is all it takes for the voice generator developed by OpenAI to clone a voice. It’s not the only system of its kind, but it shows what an AI model can do with just a few seconds of audio. Now imagine what could happen if you were one of the biggest influencers in the tech world, with hundreds or thousands of hours of published content online.
That’s what happened to Marques Brownlee.
It’s me, but it’s not me. Brownlee has 19.6 million subscribers on YouTube, 4.9 million followers on Instagram, and 6.3 million followers on X. At 30, he’s one of the most popular content creators in the tech world and one of the most recognizable faces—and voices—in the industry. He has his own way of speaking, a unique tone, and signature phrases. And, of course, thousands of hours of his voice online.
Click on the image to read the thread on Threads.
What happened? A company that makes digital business cards cloned his voice for use in an ad posted on Instagram. As Brownlee explains in a post on Threads, “There are real companies who will just use an AI-created rip of my voice to promote their product.” He says, “There’s really no repercussions for it other than being known as this scummy shady company that is willing to stoop that low to sell some product.”
The voice isn’t original, but it sounds like it is. If you listen to the video recorded by the creator himself and pay close attention, you’ll notice the voice is synthetic. Although the initial “OK” and the first sentence might sound perfect, the rest of the audio lacks intonation. It’s a flat delivery in his voice, or rather a fairly passable copy, but too flat. A more trained ear or someone interested in technology could easily detect the synthetic origin, but this isn’t always the case.
This screenshot comes from a page with 145,000 likes, dedicated solely to posting AI-generated images of grandmothers knitting. Each photo had between 600 and 1,000 likes, 60 to 200 comments, and 10 to 20 shares. Image | Xataka On
The internet has taught us this. On Facebook, some pages post AI-generated photos of people baking cakes or grandmothers knitting, and many people think they’re real, commenting and interacting. The same goes for this “restaurant” that claims to be “Austin’s No. 1 restaurant” but is actually filled with photos of AI-generated food. The account has 72,000 followers, and its images receive countless interactions.
It’s not the first time and probably won’t be the last. Jeff Geerling, another tech YouTuber and programmer, also recently had his voice cloned for use in an ad. Elecrow CEO Richard Lee apologized, stating that an inexperienced recent graduate made the video and posted it without approval.
Ukrainian YouTuber Olga Loiek was also a victim of something similar, but even more disturbing. Loiek, who studies at the University of Pennsylvania, discovered an avatar using her face and body posing as a Russian girl, posting videos on Chinese social networks. In the videos, this avatar discussed good relations between Russia and China and even claimed to be looking for a Chinese partner.
Image | Collision Conf | Xataka On
Related | I Tested ChatGPT’s Advanced Voice Mode. It’s the Beginning of Something Huge
View 0 comments