OpenAI begins releasing AI that clones voices with a sample size of just 15 seconds

OpenAI, owner of ChatGPT, has begun releasing a new content creation platform on a limited basis. What’s new is Sound EngineA service that can generate synthetic voices based on a person’s voice.

According to OpenAI, This feature already works based on 15 seconds of audio whose voice will be transcribed. This is enough to mimic someone’s tone of voice and even rhythm, but longer samples result in more complete and accurate content.

This platform has been in development since 2022 and has been partially adopted in the company’s text-to-speech APIs, but only now has OpenAI shared its practical results with the public. Voices can be heard in a post on the company’s blog.

Until now, the Voice Engine had been trained mainly to read text written from scratch in the same language. But he Can also be used to translate sentences into other languageseven preserving the original accent.

Voice-transcribing AI has benefits and risks

The idea is that Voice Engine is mainly used by information or education companies to deliver content to more people. In tests already carried out with partner developers, OpenAI has achieved: Develop a reading assistant to help children during the learning phase, including generating real-time responses on a wide range of topics.

Even those who have a condition that prevents or prevents speech you can use the platform to communicate better using synthetic voice.

Despite the excitement, the company is aware that the release needs to be gradual to catch on “How can society adapt to these new capabilities?“since “Potential misuse of synthetic sounds” in a few cases.

OpenAI will join a number of US and international organizations to discuss the responsible use of this AI. One of the rules will be prohibiting unauthorized copying of another person’s voiceIn addition Always warn the listener when speech is artificially produced.