A group of American scientists were able to train the algorithms of the neural network to recognize human presence, typical breaths between some of the words which people do spontaneously at the moment of speech production.
Researchers representing the University Carnegie Mellon is located in Pittsburgh, has published his scientific work on the website arXiv.org. Scientists have supplied the neural network incremental volume of short-term memory, allowing download into it over hundreds of hours of newscasts, in which the leading spoke English. After extensive analysis of the material, the neural network has learned to recognize the smallest phonemes in the speech of the speaker, namely, uncontrolled breaths between some words.
According to the researchers, this method could significantly advance the methods of identifying the person using speech, as the sighs of hard to control, and they are largely difficult to fake, as it is an unconditioned reflex.