[ad_1]
Artificial intelligence does not create fake photos and videos, it can also create false voices.
Until now, these voices were remarkably stable and robotic, but researchers from the startup of IA Dessa have created the most compelling voice clone we've ever heard – perfectly imitating the sound of the MMA commentator became podcaster Joe Rogan. .
Listen to excerpts from AI Rogan from Dessa below or answer a quiz on the company website to see if you can tell the difference between real Rogan and fake Rogan. (It's surprisingly difficult!)
As for the making of a convincing fake, Dessa has chosen his target well. Rogan is probably the most popular podcaster in the world and has recorded more than 1,300 episodes of The Joe Rogan experience nowadays. This provides a lot of training data for all AI systems.
The engineers of the company obviously know perfectly Rogan's favorite topics. To speculate on whether we live in a computer simulation or to admire the strength of the upper body of chimpanzees – this is Rogan's raw material.
But of course, being able to convincingly simulate someone's voice also has disturbing implications. As noted by Dessa's engineers in a blog post, cases of malicious use of fake voices include spam calls that mimic your loved ones; use false voice to intimidate or harass people; and create misinformation through fake records of politicians.
"It's clear that societal implications for technologies such as speech synthesis are enormous," writes Dessa. "And the implications will affect everyone. Poor consumers and rich consumers. Businesses and governments. "
The company notes that there are also advantages. These include creating more realistic AI assistants; faster and more accurate dubbing for television and movies; and the design of realistic and personalized synthetic voices for people with speech impairments.
We have contacted Dessa for more information about her work, but the company says that due to possible malicious uses, she will not publish her research or make her artificial intelligence models publicly available. (A position we've seen of larger AI labs, such as OpenAI, controversially retaining the final version of its AI-generating text system.)
Although it is reasonable to say that the fears about deepfakes are exaggerated (the technology has been available for years, but a counterfeit has not yet had an impact on the general policy), it is also clear that the technology will only improve and become more accessible in the world. future.
"At the moment, technical expertise, ingenuity, computational power and data are essential for the proper functioning of models like RealTalk," says the company. "But in the coming years (or even earlier), we will see the technology evolve to such an extent that a few seconds of audio is needed to create a replica of anyone's voice on the planet."
Listening to the AI, Joe Rogan talks about chimpanzees that are tearing your bullets n 'is, strangely, only the beginning.
[ad_2]
Source link