A team of scientists at Oxford University has created a new artificial intelligence system that they call LipNet. They claim that this AI can lip-read better than humans, which is interesting to say the least. LipNet is capable of doing that by watching a video of a person speaking and then it matches the text to the movement of their mouth. The researchers claim that LipNet has a 93 percent accuracy ratio which is definitely something to write home about.
However, they recognize that the system need to be tested in real-life situations before it can be ascertained just how good it is. They also suggest that automating this process is going to help millions of people as lip-reading isn’t exactly easy for humans as well, most professionals are believed to have a 60 percent accuracy ratio at best.
“Machine lip-readers have enormous 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,” the researchers wrote in their paper.
They are training the AI to teach itself which letter corresponds with which lip movement by providing it with whole sentences. They have fed almost 29,000 videos each labeled with correct text, each video was three seconds long and had a similar grammatical pattern.
Human testers were provided the same material and while they had an error rate of 47.7 percent, the AI showed its power by only posting an error rate of 6.6 percent.