Identifying "fake news" may have just gotten more difficult.
Researchers at the University of Washington have developed an algorithm that can take an audio clip and convert it into an artificial video of someone speaking those words.
This week, the team behind the technology released a video of a synthesized President Obama that they created by pulling audio from his past speeches.
“These type of results have never been shown before,” said one of the researchers, Ira Kemelmacher-Shilzerman, in a statement.
This technology has been around for years, but past attempts yielded more robotic-looking looking results.
“If you don’t render teeth right or the chin moves at the wrong time, people can spot it right away and it’s going to look fake,” Supasorn Suwajanakorn, another researcher on the project, said in a statement.
The team at University of Washington was able to overcome this by inputting hours of footage of Obama into a special computer system called a neural network. The network then tracked what shape his mouth made depending on which sound he made. Those mouth shapes were then superimposed onto an existing video of the President’s face. This combination of tactics resulted in a more authentic simulation that takes into account Obama’s distinct mannerisms.
While the developers behind this method think it could one day be used as a more efficient alternative to video chatting (streaming audio uses less bandwidth than streaming video), critics are concerned that the potential for abuse outweighs any positive outcomes.
“It introduces a new question in viewers’ and voters’ minds as to whether what they’re watching is real or whether it’s created artificially,” said Morley Winograd, a Senior Fellow at the University of Southern California’s Annenberg Center on Communication Leadership & Policy who studies the intersection of politics and technology in the information age. “The last thing we need is more suspicion in our sources of information in our political world these days.”
The researchers claim that a new algorithm to help determine whether a video is real or not could be developed by reversing their method and feeding video rather than audio into the neural network. But whether this would be effective in actually stopping the spread of rumors has yet to be determined.
“There’s no regulatory scheme that I can think of that would be adequate for policing the internet,” said Winograd.
Despite this, the researchers have already taken some precautions to try and ensure their technology will be not be used for sinister purposes.
“We very consciously decided against going down the path of putting other people’s words into someone’s mouth,” researcher Steve Seitz said in a statement. “We’re simply taking real words that someone spoke and turning them into realistic video of that individual.””