I'd be curious to see how this works with different keyboards. They tested it on a single membrane laptop keyboard, but we don't know how it profiles the sound. Is it entirely based on the frequency of keystrokes, or on tone? Did they have to train it on that specific keyboard before it was able to identify keystrokes?
Depending on the implementation, it may pay off to have a less common keyboard with modified acoustics. They may know what every identical sounding Apple laptop sounds like, but how are they going to know about switches, keycaps, and mods?