Scientists develop a computer program that can read lips with superhuman accuracy.
A new computer program has been developed that can read a person’s lips with a higher level of accuracy than even the best-trained professional lip-readers. Scientists at the University of Oxford, which also happens to have received funding from Alphabet’s DeepMind, have developed software that can read lips correctly 93.4 per cent of the time.
A system based on artificial intelligence, LipNet is capable of comprehending a video of a person speaking, and matches it with the movement of their lips to match it with words with an unprecedented level of accuracy. According to the paper shared by the scientists, “LipNet is trained end-to-end to make speaker-independent sentence-level predictions.” It is said to have ‘enormous practical potential’ with applications ranging from improved hearing aids, silent dictation in public spaces, transcribing covert conversations, speech recognition in noisy environments, biometric identification, silent-movie processing and more.
Scientists stated that they proposed LipNet will be the first model to apply deep learning for end-to-end learning of a model, which maps sequences of image frames of a speaker’s mouth to entire sentences. This model eliminates the need to segment videos into words before predicting a sentence.
Machine lip reading is usually difficult because it requires extracting ‘spatiotemporal’ features from the video. However, recent deep learning approaches attempt to extract those features end-to-end. As with modern deep learning based automatic speech recognition (ASR), LipNet claims to be trained end-to-end, to make speaker-independent sentence-level predictions.
Most importantly, LipNet does not require hand-engineered templates of speech patterns or visuals of lip movement as the system is capable of learning and self-evolving, all the while growing better at its predictions.
The flipside, though, is that the program could also be used for mass surveillance--the program could potentially be misused to eavesdrop into public conversations when paired with a CCTV, for example. This could literally spell the end of private conversations in public places.
){kind=small}
){kind=small}
){kind=small}
){kind=small}
){kind=small}
){kind=small}
){kind=small}
){kind=small}
){kind=small}
){kind=small}
){kind=small}
){kind=small}
){kind=small}
){kind=small}
){kind=small}
){kind=small}
){kind=small}
){kind=small}
 "AI-based lip reader, Lip reading, artificial intelligence, LipNet, Machine lip reading, LipNet features, CCTV footage, CCTV spyi")
){kind=small}
){kind=small}
){kind=small}
){kind=small}
){kind=small}
)
){kind=small}
){kind=small}
)
)
)
)
)
){kind=small}
){kind=small}
){kind=small}
){kind=small}
){kind=small}