Scientists from London have succeeded in developing an artificial intelligence software system that is capable of reading lips better than humans could. A group of Oxford University researchers claims that this new innovation will be beneficial for people with hearing loss.
The AI system, dubbed "Watch, Attend and Spell" or WAS beat a human lip-reading expert after the two were made to interpret a silent video by simply reading the mouth of the person. The expert lip-reader garnered a score of 12 percent while the AI was able to catch 50 percent of the silent speech. Furthermore, the AI's mistakes were relatively insignificant. The machine simply missed the "s" at the end of some words.
According to Tech Worm, the artificial intelligence program, which was created with the help of Google's DeepMind, makes use of computer vision and machine learning. The researchers "thought" the AI system to lip-read a dataset of over 5,000 hours of videos and TV footages which were collected from six various TV programs. The AI learned over 118,000 sentences and 17,500 vocabulary words.
The Tribune stated that this impressive ability by the AI software has numerous applications and help support various developments. One prominent application of the innovation is providing help to those with hearing problems. The AI will enable those who cannot hear to navigate his or her surroundings easier than before.
The AI can also be used in conjunction with speech recognition. The AI's ability to read lips will help improve automated speech recognition which, as of late last year, has already reached human parity.
As done with the experiment, the AI can be used to dub silent films. It can also be used in dictating while in a noisy situation. Joon Son Chung of Oxford University mentioned that the AI's lip-reading ability will be useful in dictating instructions over the phone. Of course, the AI can also help professional lip-readers with their craft, as well.