Voice synthesis has come a great distance since 1978’s Speak & Spell toy, which as soon as wowed individuals with its state-of-the-art potential to learn phrases aloud utilizing an digital voice. Now, utilizing deep-learning AI models, software program can create not solely realistic-sounding voices however can even convincingly imitate existing voices utilizing small samples of audio.

Along these strains, OpenAI this week introduced Voice Engine, a text-to-speech AI mannequin for creating artificial voices primarily based on a 15-second section of recorded audio. It has offered audio samples of the Voice Engine in motion on its website.

Once a voice is cloned, a person can enter textual content into the Voice Engine and get an AI-generated voice consequence. But OpenAI isn’t able to extensively launch its expertise. The firm initially deliberate to launch a pilot program for builders to enroll in the Voice Engine API earlier this month. But after extra consideration about moral implications, the corporate determined to cut back its ambitions for now.

“In line with our approach to AI safety and our voluntary commitments, we are choosing to preview but not widely release this technology at this time,” the corporate writes. “We hope this preview of Voice Engine both underscores its potential and also motivates the need to bolster societal resilience against the challenges brought by ever more convincing generative models.”

Voice cloning tech typically isn’t significantly new—there have been several AI voice synthesis models since 2022, and the tech is lively within the open supply neighborhood with packages like OpenVoice and XTTSv2. But the concept that OpenAI is inching towards letting anybody use its explicit model of voice tech is notable. And in some methods, the corporate’s reticence to launch it totally is perhaps the larger story.

OpenAI says that advantages of its voice expertise embrace offering studying help via natural-sounding voices, enabling world attain for creators by translating content material whereas preserving native accents, supporting non-verbal people with personalised speech choices, and helping sufferers in recovering their very own voice after speech-impairing situations.

But it additionally implies that anybody with 15 seconds of somebody’s recorded voice may successfully clone it, and that has apparent implications for potential misuse. Even if OpenAI by no means extensively releases its Voice Engine, the power to clone voices has already induced bother in society via phone scams the place somebody imitates a liked one’s voice and election campaign robocalls that includes cloned voices from politicians like Joe Biden.

Also, researchers and reporters have shown that voice-cloning expertise can be utilized to interrupt into financial institution accounts that use voice authentication (similar to Chase’s Voice ID), which prompted US senator Sherrod Brown of Ohio, the chair of the US Senate Committee on Banking, Housing, and Urban Affairs, to ship a letter to the CEOs of several major banks in May 2023 to inquire concerning the safety measures banks are taking to counteract AI-powered dangers.

OpenAI acknowledges that the tech may trigger bother if broadly launched, so it is initially attempting to work round these points with a algorithm. It has been testing the expertise with a set of choose accomplice firms since final yr. For instance, video synthesis firm HeyGen has been utilizing the mannequin to translate a speaker’s voice into different languages whereas preserving the identical vocal sound.