- OpenAI released a statement on Friday about security measures for its text-to-speech model, Voice Engine.
- The voice engine produces natural-sounding voices, but some are concerned that it could be used for deepfakes.
- The technology has raised concerns among lawmakers.
For the second time in as many months, OpenAI has discussed its text-to-speech tool, reminding everyone that the tool isn't yet widely available and may never become so.
“Whether we ultimately deploy this technology widely or not, it's important that people around the world understand where it's going,” the company said in a statement on its website on Friday. “That's why we want to explain how this model works, how we're using it in research and education, and how we're implementing safety measures around it.”
Late last year, OpenAI shared its Voice Engine with a small number of external users, which takes text input and a 15-second audio clip of a human voice to “generate a natural-sounding voice that closely resembles the original speaker.” The tool can create convincing human voices in several languages.
At the time, the company said it had chosen to preview the technology but not release it broadly to “make society more resilient” against the threat of “ever-more-compelling generative modeling.”
As part of these efforts, OpenAI said it is actively working to phase out voice authentication for accessing bank accounts, explore policies to protect the use of personal voice in AI, educate the public about the risks of AI, and accelerate the development of technology to track audiovisual content to determine whether users are interacting with real or synthetic content.
But despite such efforts, fear of technology persists.
President Joe Biden's AI chief, Bruce Reed, once said voice cloning is what keeps him up at night, and the Federal Trade Commission said in March that scammers are using AI to improve their work, using voice cloning tools that make it hard to distinguish AI-generated voices from human voices.
OpenAI sought to allay those concerns in an updated statement on Friday.
The company said it will “work with U.S. and international partners, including government, media, entertainment, education and civil society, to ensure their feedback is incorporated as we build.”
The company also noted that the inclusion of its latest model, GPT4o, in its Voice Engine brings with it new threats, saying that the company is “actively red teaming GPT-4o to identify and address known and unanticipated risks across a range of areas, including social psychology, bias and fairness, and misinformation.”
Of course, the bigger question is what will happen if the technology is released to the public — and OpenAI seems prepared.
OpenAI did not immediately respond to Business Insider's request for comment.
