AI model "RealTalk" perfectly imitates speakers
The artificial intelligence (AI) system «RealTalk,» developed by a U.S. company, realistically mimics the voices of celebrities. The AI model has now imitated the voice of U.S. podcast producer Joseph Rogan almost perfectly.
"Clearly, the social implications of speech synthesis technologies are enormous," the developers said. Users can take a quiz on the website to see if they can tell the difference between the fake voice and the real Joe Rogan.
"In the area of photos and videos, we have already come so far that manipulations can hardly be detected. The voice is just another step in this direction, which, in addition to positive applications, could also be misused," describes Werner Purgathofer from the TU Vienna's Visual Computing & Human-Centered Technology research department in an interview with the Pressetext news agency. Even the media would then find it difficult to determine what is original and what is fake.
Despite the risks, RealTalk manufacturers emphasize the positive aspects of such developments. These include more realistic AI assistants, faster and more accurate dubbing for TV and movies, and personalized support for people with speech disorders. In terms of detailed information about the technology, the company is keeping mum. It does not want to publish the research in order to prevent potential misuse.
"Right now, technical expertise, computing power, as well as data are required to make models like RealTalk work well," the company says. "In the coming years, we will see technology advance to the point where only a few seconds of audio material is needed to create a lifelike replica of anyone's voice," the developers add. (pte)
