OpenAI announced models for converting text into audio Voice Engine. To clone a person’s voice, the neural network only needs a 15-second sample. says on the company’s blog.
OpenAI’s development differs from its competitors in that, as an example of a neural network, you simply provide a brief 15-second sample of a voice message from the person whose voice you want to transcribe and enter the text you want to voice. in the request field.
OpenAI experts propose using this technology to restore voice in people suffering from voice loss, and also name potential threats posed by modern generative models and ways to solve them. For example, banks and other organizations are advised to phase out voice-based biometric authentication, as well as educate people about the threats of AI and how to distinguish generated content from real content.
OpenAI also said that the development of the Voice Engine started at the end of 2022 and that both licensed and open data were used in the model training process. It is unknown when the service will be available to the public.
Previously “Yandex” educated Neural networks to protect internet users from fraudsters.