Are you sure that this is really a big problem problem in the up to 50 kHz range? The plot at the bottom of
this page suggests that the magnitude of attenuation due to humidity would be not greater than 0.5 dB/ft in this range. The listetning distance will be probably about 1 m.
Well it's certainly what they did with the scale models, but you have to bear in mind that they were trying to make accurate scale measurements from them, and they were trying to minimise the losses over several reflections from absorptive material.
At 1m, you're only just going to get a coherent audio image from most 2-unit speakers - things generally improve at about 5-6 feet away, depending somewhat on the environment. How flat the response of any of these speakers
actually is out to 40kHz is also another moot point - just claiming a response out to 50kHz hardly matters in quantative terms, because nobody is going to be able to measure it accurately and predictably anyway. There is a long history of these sorts of measurements changing every time they are made, and every time they are made by a different person, and you don't get any confidence at all about the results until you've collated
loads of readings, in just the same way that you have to with reverberation measurements.
I have a seal leak detecting system that runs at about 40kHz, and the detector for this does a direct hetrodyne translation back down to audio frequencies. The level variations you get from this by waving the rather directional reciever around in a room with it are considerable, but in an anechoic space, whilst they certainly become more predicatable, you couldn't exactly claim that they were
that much improved... you really do have to be
pretty careful about any claims you make about ultrasonic propagation.