Meta shows text-to-speech AI that can convert text to audio – IT Pro – News

Meta has proven a text-to-speech program that permits customers to transform written textual content to audio. Voicebox works in six languages, together with French and German, however Dutch doesn’t work but. The software won’t be made public in the meanwhile to forestall abuse.

Meta say that Voicebox is a generative AI that may create audio recordsdata from textual content. In response to Meta, this solely requires a chunk of audio of at the very least two seconds. Voicebox can then edit the textual content itself in six languages. Along with English, these are additionally French, German, Spanish, Polish and Portuguese.

Voicebox can even edit an audio message wherein a textual content is spoken by itself. For instance, the software can appropriate mispronounced phrases or filter out background sounds comparable to a barking canine.

Meta has one stream matchingmannequin used to make the textual content sound pure. Stream matching is an AI coaching mannequin that Meta designed itself, which is predicated on steady normalizing flows. In a analysis paper Meta says the mannequin has been skilled on 50,000 hours of audio in every of the six supported languages. The mannequin would have an error fee of just one.9 % in spoken phrases.

Meta won’t disclose both the software or the underlying mannequin in the meanwhile. The corporate says such a software has “potential to be misused and damage folks.” That’s the reason it solely needs to publish an method and the ends in a scientific paper, however not the software itself. Meta doesn’t say whether or not that may occur sooner or later. The corporate does put some demos on-line in which you’ll be able to hear examples of the AI.