OpenAI’s voice cloning AI model only needs a 15-second sample to work

OpenAI has introduced Voice Engine, a text-to-voice generation platform that can create a synthetic voice based on a short sample of someone’s voice. This AI-powered technology is being provided to select companies such as Age of Learning, HeyGen, Dimagi, Livox, and Lifespan for various applications including education, health, and communication.

The Voice Engine technology has been in development since late 2022 and has already been used to power preset voices for text-to-speech APIs and ChatGPT’s Read Aloud feature. OpenAI has limited access to the model to around 10 developers and emphasizes the importance of using the technology responsibly and ethically.

The growing field of AI text-to-audio generation is evolving, with a focus on voice generation presenting new challenges and opportunities. Companies like Podcastle and ElevenLabs are also exploring AI voice cloning technology. However, concerns around the unethical use of AI voice technology, such as robocalls using AI voices, have led to regulatory action by the US government.

To mitigate potential risks associated with Voice Engine and similar tools, OpenAI has implemented usage policies that prohibit impersonation without consent, require explicit consent from the original speaker, and mandate disclosure that voices are AI-generated. Additionally, watermarks have been added to audio clips for traceability, and monitoring systems are in place to track usage.

OpenAI proposes steps to address concerns related to AI-generated voices, including discontinuing voice-based authentication for sensitive accounts, establishing policies to safeguard individuals’ voices in AI, increasing awareness about AI deepfakes, and developing tracking mechanisms for AI content. By implementing ethical safeguards and responsible practices, OpenAI aims to ensure that Voice Engine and similar technologies are used for positive applications across various industries.