It's kinda stupid, and once they put in echo cancellation it won't work anymore.
The idea is, to record high quality sound commands, spam them out or tie them to websites or something, hope the person who gets hit with the sound has a mic, that it's on, that their speakers are on and loud enough for the mic to hear but not too loud that the sound doesn't play clearly, and that they have voice commands on, and the victim has to be AFK or not get suspicious when he hears "Start........Run........etc", and even then the attacker could only run software on the computer, and is limited by what voice commands allows you to do, so the worst they can do is randomly delete stuff.