It’s hard to even pinpoint what a new algorithm under development is going to be used for, but it seems exciting to develop artificial intelligence capable of filtering out certain words in bulk. Research in Google Research Blogged by Inbar Mosseri and Oran Lang, Google’s two software engineers.
The starting point was that while people are able to filter out, isolate, and understand the speeches by a certain speaker even if they speak more than once, computers are not able to do so at this time. This is an ever-increasing problem today when the various machine assistants are spreading.
The underdeveloped algorithm developed by Google researchers can solve this problem by watching the face of the speaker, associating visual signals with the sounds. Software engineers began to develop the system by first learning the system of recognizing the individual speech of some people, then bringing together a virtual party, which included several speakers at the same time, with background noise. In this environment, they continued their AI training, focusing on faces to isolate individual speakers and filter only their audio tracks. This was successful even if the speaker partially covered his face with his hand or microphone.
The system is likely to be able to build into Google Video Chat Hangouts (Hangouts, Duo) sometime, but it may also be useful for filming and tuning to hearing aids. And, of course, there may be a wide range of illegal or secret service applications as well, a new generation of listening devices may appear with such algorithms.