Free Trend Report Free 2018 Report & eBook

Get the top 100 trends happening right NOW -- plus a FREE copy of our award-winning book.

Keynotes.org Our Research Methodology

This article is one of 350,000 experiments. We use crowd filtering, big data and AI to identify insights.

Oxford's 'LipNet' Can Read Lips with Over 90 Percent Accuracy

 - Nov 9, 2016
References: arxiv.org & gizmodo
Researchers from the University of Oxford's Computer Science Department have developed a software program called 'LipNet' that can read people's lips with incredible accuracy. The program can take a video sequence as input and analyze it frame by frame in order to predict complete sentences.

Lipreading is a difficult task that the average person can't do whatsoever. People with experience lipreading -- such as those who are hard of hearing -- fare much better, but still have an average performance of just 52 percent accuracy. In Oxford's tests, LipNet performed with 93 percent accuracy -- an incredible improvement above humans.

LipNet was trained using GRID, a data set of 64,000 English sentences. The researchers plan to adapt the system for real-world sentences in the future.