by Neil Bauman, Ph.D.
A new study by the University of East Anglia has revealed the surprising results that computers are now better at lip-reading (speechreading) than humans are. In one demonstration. the computers recognized 80% of the words presented, whereas the 19 human speechreaders only recognized 32% of the words. (1) That is a remarkably significant difference.
Even more surprising is the fact that with just four hours of training, the computers helped the human speechreaders markedly improve their lip-reading skills. This new research opens the way for teaching hard of hearing people improved speechreading skills that will improve their accuracy dramatically.
Looking into the future, I can foresee the day when a hard of hearing or deaf person, using a video camera attached to their laptop will be able to talk with a hearing person via their laptop. The laptop will speechread the person and print what they say on the screen with a reasonable degree of accuracy.
Unlike speech recognition software, which listens to the person speaking and converts their speech to text, this speechreading software converts facial (mouth) movements to text. Thus, it works just as well in noisy places as in quiet situations, whereas the speech to text software requires relatively quiet situations so that it can hear human speech over the noise.