Communications of the ACM
Is No Secret Safe? Lipreading Robot Proves More Accurate Than a Human in Deciphering Speech
Researchers say a new program is more accurate at reading people's lips than many human experts.
Credit: Yannis Assael
Researchers from the University of Oxford in the U.K. have developed LipNet, a new program they say is more accurate at reading lips than human experts.
The researchers found LipNet can determine what people are saying by reading their lips 93.4% of the time, while the average accuracy of an experienced lipreader is about 52%.
LipNet uses a neural network to map mouth movements of people to a database of set sentences. The researchers trained LipNet with nearly 29,000 videos of two men and two women giving a strange series of commands, with cryptic phrases such as "set blue by A four please." LipNet learned to match the movements of people's mouths with the known commands by analyzing each individual video frame.
Going forward, the researchers want to train LipNet with more real-world examples.
"Machine lipreaders have enormous practical potential, with applications in improved hearing aids, silent dictation in public spaces, covert conversations, speech recognition in noisy environments, biometric identification, and silent-movie processing," says University of Oxford doctoral researcher Yannis Assael.
From Daily Mail (United Kingdom)
View Full Article
Abstracts Copyright © 2016 Information Inc., Bethesda, Maryland, USA
No entries found