The Ability to Mimic Voices with AI Comes with Caution

OpenAI has developed a Voice Engine capable of accurately replicating any individual’s voice with a mere 15 seconds of audio. This technology, initially crafted in 2022 and integral to ChatGPT’s text-to-speech functionality, has been withheld from public release due to concerns over its potential misuse, especially in spreading misinformation around election times.

OpenAI stresses the importance of societal preparedness against the risks, advocating for the discontinuation of voice-based authentication and the implementation of robust policies to safeguard individual voices. The technology features a watermarking system to ensure the original speaker’s consent, setting a precedent in ethical AI use. Meanwhile, competitors like ElevenLabs are emerging with similar technologies, emphasizing protective measures against misuse in sensitive areas such as politics.

What we know so far

🤖 Initially developed in 2022, used in ChatGPT’s text-to-speech feature.