SEATTLE, WA – Blurring the line between fact and fiction just got easier thanks to a new technology devised by computer scientists from the University of Washington.
Using 17 hours of Barack Obama’s weekly video addresses, UW scientists created numerous computer-generated mouth shapes based on the former US president’s audio clips. Digitally rendering these shapes with photorealistic quality, they restored them with the rest of Obama’s facial expressions. The video was ultimately retimed, allowing the audio track to move in sync with Obama’s body movements.
Although bonafide audio were the only recordings used for the study, the team was able to skip and reorder Obama’s sentences seamlessly – even using audio from an Obama impersonator for a perfectly authentic-looking video. The evolution of voice-synthesis software continues to roll out simple solutions for compelling, fake audio.
The video technique is still dependent on the amount and quality of footage a subject has available. Several hours of footage are needed for the technique to work at minimum. Obama was chosen for his consistent tone of voice, high-definition close-ups, and publicly accessible video footage.
Similar studies have successfully altered recorded facial expressions, but the UW’s study is the first to generate video directly from audio clips. Public figures who lack sufficient video or audio resources would be much harder to duplicate – all the while giving computer scientists time to develop software to detect falsified video.
Such technology has plenty of benign applications – from helping Hollywood render computer-generated characters to helping hearing-impaired people lip-read audio during a phone call. The potential for generating fake news stories and falsifying video testimony is a grave concern for the public at large.
In the era of fake news, the burden unfortunately falls to citizens to educate themselves on these rapidly advancing technologies and hold leaders accountable – in their own words.
UW paper: http://grail.cs.washington.edu/projects/AudioToObama/
Stanford research: http://www.graphics.stanford.edu/~niessner/thies2016face.html
Falsified face tips: https://www.youtube.com/watch?v=I3l4XLZ59iw
Economist – fake video/audio studies: https://www.economist.com/news/science-and-technology/21724370-generating-convincing-audio-and-video-fake-events-fake-news-you-aint-seen