OpenAI's voice cloning breakthrough allows synthesis from just 15 sec of speech

OpenAI has unveiled a voice cloning AI model named Voice Generation, which can generate synthetic voices with remarkable accuracy using only a 15-second sample of someone's voice.

Developed since late 2022, Voice Generation powers the Read Aloud feature in ChatGPT, providing a seamless audio experience for users. Limited access to the text-to-voice generation platform, Voice Engine, has been granted to select companies, including Age of Learning, HeyGen, Dimagi, Livox, and Lifespan.

Trained on a mix of licensed and publicly available data, the model offers a versatile solution for generating lifelike voices. However, OpenAI is cautious about its usage, implementing strict policies to prevent misuse, including impersonation without consent and unauthorized voice cloning.

To mitigate risks associated with AI-generated content, OpenAI advocates for phasing out voice-based authentication for sensitive accounts, implementing protective policies, educating users on AI deepfakes, and developing tracking systems for AI content. These measures aim to ensure responsible and ethical use of AI voice cloning technology in various industries.