Machine learning has led to a new artificial intelligence (AI) program that outperformed professional lip readers and the best AI to date, Science magazine reports.
The program performs with just half the error rate of the previous best algorithm, and could put lip reading in the palm of everyone’s hands.
The researchers took 140,000 hours of YouTube videos and designed a program that created clips a few seconds long with the mouth movement for each phoneme, or word sound. After the program filtered out non-English speech, non-speaking faces, low-quality video, and video that wasn’t shot straight ahead, they cropped the videos around the mouth, yielding nearly 4000 hours of footage that included more than 127,000 English words.
Integrating the program into a phone would allow deaf people to take a “translator” with them wherever they go.
Read the full story: