After text, images and videos, OpenAI is tackling sound, with Voice Engine, a new AI tool capable of cloning human voices with striking realism. Another bulky technology, but one that could be misused.
Once again, artificial intelligence has made it possible to develop a tool that is as spectacular as it is frightening. Last Friday, April 26, the Californian start-up OpenAI, known for having developed ChatGPT, a conversational agent, and Sora, an AI capable of generating breathtakingly realistic videos, unveiled a brand new tool capable of cloning voices . Called Voice Engine, this technology boosted by artificial intelligence can in fact reproduce a person’s voice after listening to it for just fifteen seconds. This system is therefore not entirely new, since OpenAI had already developed a first version at the end of 2022, which it had integrated with the speech synthesis functions in ChatGPT. However, this is the very first time that the start-up has publicly communicated about this new service offering.
Voice Engine: perfect imitations… and dangerous!
In any case, the results published by the company are stunning. Indeed, Voice Engine is capable of reproducing accents, intonations, nuances and, more generally, all the specific characteristics of a person’s voice. Thus, it is increasingly difficult to tell the difference between a cloned voice and a real voice. Thanks to this technology, OpenAI intends to offer new possibilities to its users. For example, the tool can be used as a reading aid for people who have lost their voice, a content translator (since it works in French and other languages), or even a reading assistant for children or people who cannot read. But that’s not all ! Artificial intelligence may also be introduced into the daily lives of literature fans. Indeed, Voice Engine could be used in the design of audio podcasts, audiobooks, and even on the radio.
Unfortunately, the technology has already been used for criminal purposes. While the United States is in the midst of an election campaign, cybercriminals have spoofed Joe Biden’s voice to call voters. Faced with such danger, the company took a deliberately cautious stance and took several safety measures. First of all, Voice Engine is not accessible to the general public, but only to partners who have accepted OpenAI’s conditions of use, prohibiting them from practicing identity theft or exploiting a voice without the consent of the person.
Partners are also obliged to indicate that the voice used was generated by artificial intelligence. “NOTWe work with U.S. and international partners from government, media, entertainment, education, civil society and other sectors, and we take their feedback into account as we develop the tool“, indicated the company, before concluding: “We have implemented a range of security measures, including a watermark to be able to trace the origin of all audio generated by Voice Engine, as well as proactive monitoring of its use.”