Recently it became known that the new Vall-E neural network from Microsoft is able to fake the voice of a particular person up to intonations. About this
Vall-E taught the Librilight library containing 60 thousand hours of English-language speech from more than 7 thousand people. For the system, enough recording lasts three seconds to get a high -quality fake.
The publication notes that the voice is imitated very reliably, it retains the timbre and the correct emotional coloring of the original.
According to Microsoft, Vall-E will not be distributed in the publication for security reasons so that scammers do not use the neural network.
/Media reports cited above.