OpenAI announced a pre-release of It is the Engine, a model of text-to-speech, which can produce a sound synthetic reflection of the introduction to the text and a sample sound for 15 seconds. The technology, developed by the end of the year is 2022, and used for the blog and Read Aloud to the ChatGPT, it can be tested by only a limited number of partners.
In only 15 seconds of audio to clone a voice in your
While the Engine was training with a mix of public record and, to be licensed (OpenAI did not give the details). The model generates a sound almost identical to that of the original, as it takes a sample from the audio, the 15-second interval as well as the introduction.
It can also be used for the application of the material. Some of the examples provided by the partners, have been published in the official blog. The Age of Learning to create a tool to help with reading, while the HeyGen used While the Engine to translate the audion from English to other languages (Spanish, French, German, japanese, and mandarin).
These models also can be used to generate the falsification of the depth to the audio. OpenAI is aware of the risks involved, and therefore It is the Engine actually has been tested by some of the partners are selected who are required to follow strict rules. It is prohibited to use the template to clone a voice, without the consent of a clear and informed consent of the speaker to the original.
The partners must also disclose that the noise is generated by the intelligent man, and must implement a set of measures to track the origin of each sound to be generated (for example, by filigranit). Some of the issues of privacy, security and safety need to be solved before the introduction of the general.
Staying on the theme, Microsoft and OpenAI said that they had planned to build in the center of the data, which will be the host of a superkompjuter IT's called the Stargate. According to the sources of The Information, the cost is about 100 billion dollars.
Discussion about this post