Date:12/12/17

Neural network can now recognize a person by his breath

A group of American scientists from the University of Carnegie Mellon (Carnegie Mellon University) claim that they were able to teach the computer to identify people by their breaths. This possibility was justified by the fact that the breaths are difficult to control, and this completely excludes the possibility of their forgery.
 
The research involved a group of people with the help of whom a method based on super-vector traits included in many voice recognition systems was tested. In this case, the spectral images of temporal fragments of speech are transformed into the shape of a vector pattern. After preparatory actions, artificial intelligence searches for regions of condensation with their transformation into the prototype of the supervector, which is analyzed using different classifiers. At a certain stage, scientists were asked to tune the neural network to recognize only the breaths.
 
In order to test this method, scientists used a set of available data collected on the basis of 100 hour newscasts in English. The recordings covered the period from 1997 to 1998. The news were used for training  the  artificial intelligence, and only sounds between words were used. As a result, the computer recognized the identity of the person under the previous method with a precision of 72 - 74 percent, and the new methodology increased the ability by 91 percent, DNI24 reports.

Views: 312

©ictnews.az. All rights reserved.