US firm develops generative AI capable of reproducing a human voice

US venture firm OpenAI says it has developed generative artificial intelligence technology that can recreate human speech by using a short recorded voice sample of the original speaker.

OpenAI revealed its newly-developed AI "Voice Engine" on Friday.

The company says "Voice Engine" uses a 15-second audio sample and inputted text to generate human speech. "Voice Engine" reads the inputted text in a voice that closely resembles the original speaker's.

The firm says the new model can also translate the AI-generated speech into foreign languages. The model can reportedly preserve the native accent of the original speaker as it does so.

OpenAI says the AI technology can be used to support people who have trouble speaking, due to disease or other reasons. The firm says businesses can also use the AI technology to transmit information to foreign countries.

But there is concern that this kind of chatbot could be used to spread fake information during election campaigns or in other situations.

OpenAI said, "We recognize that generating speech that resembles people's voices has serious risks, which are especially top of mind in an election year."

The company says it does not plan to widely release the technology for the time being, because of the possible risks.
It says it still needs to take steps to prevent the technology from being misused. It cited the generation of the voices of celebrities as one type of misuse.