On Friday, OpenAI unveiled a voice-cloning tool, which they intend to regulate closely until adequate measures are established to prevent the creation of deceptive audio clips.
The tool, named “Voice Engine,” has the capability to replicate an individual’s voice from just a 15-second audio snippet, as detailed in a blog post by OpenAI that discussed a limited test of the technology.
Acknowledging the potential dangers, particularly in an election year, the company emphasized, “The ability to generate speech mimicking individual voices carries significant risks,” highlighting their cautious approach to its deployment.
OpenAI stated, “We are collaborating with partners in the U.S. and globally from various sectors including government, media, entertainment, education, and civil society to integrate their insights into our development process.”
Researchers concerned with disinformation are alarmed by the potential widespread abuse of AI-powered tools, especially voice cloning technologies, in a crucial election year. These tools, being affordable, user-friendly, and difficult to track, pose a significant risk of misuse.
OpenAI has expressed a commitment to a deliberate and well-informed strategy regarding the wider distribution of their voice cloning tool, citing concerns over the possibility of its misuse in creating synthetic voices.
This careful approach follows an incident where a political consultant associated with a Democratic contender against Joe Biden was found responsible for a robocall that mimicked the President’s voice. Engineered by a strategist for Congressman Dean Phillips of Minnesota, the AI-generated call featured a voice resembling Biden’s, discouraging voters from participating in New Hampshire’s January primary.
The incident heightened concerns among experts about the potential flood of AI-generated deepfake disinformation in the 2024 U.S. presidential election and other critical elections worldwide this year.
OpenAI has stipulated that partners experimenting with the Voice Engine adhere to guidelines that mandate clear and informed consent from individuals whose voices are replicated. Additionally, it’s essential that audiences are made aware when the voices they listen to are AI-generated.
To enhance safety, OpenAI explained, “We’ve introduced safety protocols, including watermarking to identify the source of any audio produced by Voice Engine, and proactive surveillance of its usage.”
