Sogou company held a media communication in Beijing, formally launched a new human-computer interaction technology "lip recognition", the technology can be recognized by the machine vision, do not listen to the sound, just by identifying the speaker lip action, you can interpret the speaker said content.
It is understood that lip language recognition is a technology based on machine vision and natural language processing in one, Sogou end-to-end deep neural network technology for Chinese lip sequence modeling, through thousands of hours of real lip language data training, the official Said that in the non-specific open spoken language test set, lip recognition system can achieve more than 60% accuracy, in vertical scenes such as car, smart home and other scenarios can achieve 90% accuracy.
At the application level, field engineers introduce that lip recognition can assist in voice interaction and image recognition. For example, in the car scene, when the surrounding noise is too large, it can interfere with voice commands and can be avoided through lip-recognition technology. Of the public places to ensure the privacy of speaking; in the field of security, because most of the current monitoring camera only without Mike, lip recognition can help public security access to important speech information; lip recognition can also help congenital hearing-impaired people or the elderly.