The original article http://sandlab.cs.uchicago.edu... has some interesting graphs showing efficacy under certain situations. Looks like sometimes it can be fairly effective. What's missing are the frequency (ies) and sound energy the bracelet (transducers) puts out. Normal conversation is about 60 dB in amplitude. If you can get an SPL out of the transducers exceeding this, then the source (conversation) to noise (jamming) ratio will be low enough that the level will swamp the input of the listening device. The signal has to go through an A/D converter on the front end, and if you can get enough volume out of the jammer, that's all it will pick up. No amount of post filtering will make the signal intelligible.
Why is the mic picking up ultrasonics? In any case, to defeat this they could just put an analog low pass before the ADC. There should be one there already (the anti-aliasing filter for the ADC). Band-limiting the input to around 4kHz like POTS should do the trick. Not that I want this jammer to be defeated...
Ocean: A body of water occupying about two-thirds of a world made for man -- who has no gills. -- Ambrose Bierce