Various artificial intelligence systems already know how to do things that it is time to think about what the plot of the Terminator franchise is not so fantastic. Moreover, every day AI learns new techniques. For example, recently Google’s specialists have trained their artificial intelligence to identify voices of individual people in the crowd noise with amazing accuracy.
For starters, Google engineers have created a technology based on machine learning which brings out the voices of individuals due to “the effect of party.” This is something that can do all of us: talking with a person surrounded by extraneous noise, we can understand what he says, analyzing not only speech, but also facial expressions, filtering out the “disturbing” sounds. AI Google works in a similar way: through analysis of the mimic muscles of the face and all the noise, he distinguishes the voices of different people and creates a separate audio track. Then Google taught artificial intelligence to recognize people’s faces, to isolate their voices and suppress background noise.
Therefore, the system can distinguish from the crowd of noisy people right person and to recognize his voice in real-time. This was possible due to the fact that Google engineers are “trained” the AI on more than 100 000 video lectures and speaking videos from YouTube with background noise. The new development, according to Google, can be used, for example, to improve the quality of video chat to during group communication, the user can focus on a single source. The result of the algorithm Google you can see on the video, available below.
Artificial intelligence, Google has learned a new trick
Vladimir Kuznetsov