Artificial intelligence has learned to single out one person's speech from the general noise emitted by a multitude of talking people

People who often attend various receptions and parties know that not a lot of attention is required in order to concentrate on the speeches of the only person speaking in a room filled with the speech of a large number of people who are there. All the most complicated work on the allocation of the speech of the interlocutor we are interested in is taken up by our brain, but when it comes to technologies for automatically singling out one person's speech from the general noise, the results of such technologies leave much to be desired.

Breakthrough in this area was made by specialists of the Japanese company Mitsubishi Electric. At the R & D Open House event, which took place in Tokyo on May 24, a new technology for voice separation and recognition was introduced to the public. This technology was able to successfully separate the speech of two people speaking the same microphone in the same sentence in different languages. The separation system was practically run in real time, with the delay not exceeding 3 seconds. The accuracy of the division of speech was 90 percent. But when three people started talking at the same time at the same time, the accuracy of the division of speech dropped to 80 percent.

These tests of the speech separation system were carried out under ideal conditions, in a room with good sound insulation, provided that silence was observed by other people present there. But in this case the new system showed the highest results, the accuracy of other similar systems under the same conditions does not exceed 50 percent.

dailytechinfo.org 

© All rights reserved. Citing to www.ict.az is necessary upon using news