13th November 2016
AI can beat humans at lip-reading
The University of Oxford has demonstrated "LipNet", a new AI algorithm capable of lip-reading over 40% more accurately than a real person.
2016 has been a big year for artificial intelligence, with many important breakthroughs that we've covered on our blog. Yet again, what was once confined to science fiction has become a reality, as this week a research team presented a new AI lip-reading system able to beat humans.
The University of Oxford's Department of Computer Science has developed "LipNet", a visual recognition system that can process whole sentences and learn which letter corresponds to the slightest mouth movement.
"The end-to-end model eliminates the need to segment videos into words before predicting a sentence," the research team explains. "LipNet requires neither hand-engineered spatiotemporal visual features, nor a separately-trained sequence model."
While an experienced human lip-reader can achieve accuracy of 52%, the LipNet efficiency is 93%. It's eerily reminiscent of HAL 9000, the sentient computer in Arthur C. Clarke's 2001: A Space Odyssey.
However, while LipNet has proven to be very promising, it is still at a relatively early stage of development. So far, it has been trained and tested on short, formulaic videos that show a well-lit person face-on. In its current form, LipNet could not be used on more challenging video footage – so it is currently unsuitable for use as a surveillance tool. But the team is keen to develop it further in real-world situations, especially as an aid for people with hearing disabilities.
• Follow us on Twitter
• Follow us on Facebook