Facebook AI: how to separate two overlapping items

The Facebook AI team published the results of a research aimed at analyzing two overlapping voice audio streams in order to distinguish them to make them two separate entities. What the human ear is able to do through an effort of sensory focus, Artificial Intelligence can do it with much greater difficulty, but with far better results in perspective thanks to the many possible fields of application.

Thus the AI ​​separates two overlapping items

The project involves the use of specific data sets capable of greatly exceeding the current state of the art: thanks to the model put in place by the Facebook laboratories, only it is easier to distinguish two overlapping voices (for example in a phone call, in a dialogue or in a natural environment), but it is also possible to eliminate the background noise and ensure that only the desired spoken flow can emerge:



The fields of application, as evident, are many. For example, useful techniques could be developed to improve listening by users who suffer from hearing loss, as well as digital tools to understand audio instructions imparted without confusion between multiple spoken streams. A voice assistant would therefore become more precise, just as a recording tool could better distinguish the voice of a single user among many other people. You can also improve voice-text transcriptions, you can get better refined subtitles and you can accelerate on a multitude of applications based on listening and interpreting voice audio streams.

What this might mean for Facebook is easy to imagine, as it would be to turn into classified information and catalogabili what up to now was only confusion (so, in the video upload, so in the audio tracks sent). Separate multiple entries with a reliable process is therefore an objective that is as complex as now reach: Facebook has shown that Artificial Intelligence can succeed where the man could already, but where only the machine will be able to automate any mechanism for propagating the extreme of the fields of application as possible.

Source: Facebook TO