Voice deepfakes, anyone?

I’ve beomce more an dmore fascinated with deepfakes, both faces and voices. But for some voice experiments I want to run (making interesting voices for small background characters), I have been unable to find actual software to do it. Those amusing videos on Youtube never seem to say what they use to make those fake voices of Obama or Arnie or whatever. Does anyone know where to get software like that? Not an API, I have enough on my tech table, just an out-of-the-box pice of software or online feature or the like?

1 Like

Aside from a promising demo back in 2016, Adobe’s Voco software appears to be a tech demo that may never be released: https://youtu.be/I3l4XLZ59iw

This Two Minute Papers video has a link to a paper called Neural Voice Puppetry and online demo in the description: https://youtu.be/VQgYPv8tb6A

But I’ll drop them here for you:
Paper: https://justusthies.github.io/posts/neural-voice-puppetry/
Demo: http://kaldir.vc.in.tum.de:9000/

Sadly, they seem to have discontinued that one :frowning:

1 Like

Just came across this one:

Apparently, Overdub is a part of the Descript application that lets you edit recorded audio by synthesizing new audio based off input text and a voice model trained from randomly generated phrases. This is supposed to make it more “deep fake-proof” where you can’t just train it on just anyone’s voice.

1 Like

Ohhhhh… new toys!!!

1 Like