Synthesizing Obama: Learning Lip Sync from Audio

I ran across this tonight, it's amazing and beyond creepy at the same time. I think back to the alleged fake Julian Assange videos (the collar one) and the new software that can mimic anyone's voice with only a one minute clip.

Imagine watching a video of yourself confessing to a crime and having to prove in court that the video confession is fake. What if a propaganda video of a world leader came out declaring war and shtf before the video gets debunked. Advertisers would just have to buy image rights to celebrities and could make them say anything like. This is scary.. awesome but scary.

Given audio of President Barack Obama, we synthesize a high quality video of him speaking with accurate lip sync, composited into a target video clip. Trained on many hours of his weekly address footage, a recurrent neural network learns the mapping from raw audio features to mouth shapes. 

grail.cs.washington.edu...


www.theverge.com...

originally posted by: PlasticWizard

Imagine watching a video of yourself confessing to a crime and having to prove in court that the video confession is fake.

This would show up quite easily on image forensic analysis, I would imagine.

a reply to: GetHyped

You forget, when it's politics, no one bothers with evidence.

a reply to: PlasticWizard

Oh I can see how editing out words, etc...if there was an agenda behind a certain speech and the elite wanted to pull strings on the puppet (president)

I could also see this being used in the movie industry where if an actor/actress messes up their words in an expensive scene they can easily do a voice over.

Cool find OP. S&F

Technology and The Science of Deception shall be another book I write some day.