Google's AI in collaboration with DeepMind can lip-read better than humans

Utilizing its high-end Artificial Intelligence, Google is currently gearing up to introduce its own lip-reading software. The software would be made in partnership with the University of Oxford and Google’s DeepMind division.

The software, under progress, made use of up to five thousand hours of BBC footage to get acquainted with the system and to create its own neural network.

In result, a program known as Watch, Listen, Attend and Spell (WLAS) has been developed to achieve the accuracy of 46.8 percent while watching a video. This may not seem like a much reliable percentage, but when compared to the human experts who have an estimated experience of 10 years; they could only get words correct 12.4 percent of the time.

The lip-reading software was trained with a dataset of more than a hundred thousand natural sentences, along with the marginal note of 200 plus videos; and after hours of training it has reached this stage.

The lip-reading AI would be rolled around the world, once the software is perfected.