OpenAI Unveils AI Voice Engine That Clones Your Voice From A 15-Second Sample
OpenAI has unveiled the first results of testing a new feature that can read texts aloud in a convincing human voice. This opens a new era in the development of artificial intelligence and increases the likelihood of deepfakes.
The company showed off early versions and use cases of a text-to-speech feature called Voice Engine in a short demo that was available to about 10 developers. However, OpenAI decided to delay the widespread rollout of the feature, and they told reporters earlier this month.
Since last year OpenAI has offered to test Voice Engine with a small group of partners – and received examples of use:
- Helping struggling readers and children through natural, emotional and diverse voices;
- Translating content such as videos and podcasts will allow authors and companies to expand their audiences around the world by communicating information in their own and employee voices;
- Support for people with speech problems, therapeutic applications, educational supplements;
- Help in restoring the voice in cases of sudden and degenerative speech disorders;
OpenAI’s Voice Engine differs from the company’s previous audio content efforts in that it can create compelling voices with the unique intonation of different people based on just 15 seconds of audio recording of the speaker. During a demo where the AI spoke through the voice of OpenAI CEO Sam Altman about the technology itself. The speech was indistinguishable from Altman’s real voice, but was entirely created by artificial intelligence.
