By: Rohit Nagar
Microsoft made a new AI speech generator called VALL-E 2 that sounds so real it can't be released to the public.
VALL-E 2 can copy a person's voice using just a few seconds of audio.
Researchers say VALL-E 2 creates speech that is as good as human speech.
It uses two special methods, "Repetition Aware Sampling" and "Grouped Code Modeling," to make speech sound natural and quick.
VALL-E 2 was tested using audio samples and proved better than past AI systems at matching human speech.
The AI is not being released because it could be misused, like for faking voices.
Microsoft is keeping VALL-E 2 as a research project and has no plans to make it public.
In the future, this AI could be used for education, entertainment, and other helpful applications.