OpenAI’s new AI can clone a person’s voice from 15-second audio

OpenAI has announced that it is ready to introduce a new AI tool Voice Engine. It can be used to clone people’s voices based on audio recordings of their speech lasting only 15 seconds. In this case, the generated voices sound not only natural, but also emotional and realistic. Work on the technology has been underway since 2022.

The company said that they believe the technology is useful. It can be a great solution to help with reading and translations. In addition, Voice Engine looks like a panacea for everyone who suffers from degenerative speech disorders. As an example, the developers talk about how the tool has already helped one such patient successfully complete a school project. But the technology also raises significant concerns. OpenAI points to the high risks associated with its widespread adoption.

Fraudsters are likely to actively exploit the capabilities of Voice Engine or its analogs to their advantage. To reduce the risks, the company has established several rules that must be strictly observed by all who intend to use the service. First of all, it is required to notify the audience that the voice is created by a neural network. Watermarks and a proactive monitoring system are also provided. There will also be a ban on cloning the voices of famous personalities. And, of course, no one is allowed to use someone’s voice without their consent.

There is no exact date when Voice Engine will begin rolling out yet – but the subscription fee is known. It will be $15 for 1,000,000 voiced characters – comparable to a full-fledged book. A subscription to the HD version of the service for $30 is also expected, but its benefits are not yet known.

Read also: An ordinary Android smartphone can pick almost any lock in a hotel room