Suppose you set up a high-Q acoustic resonator strongly coupled to the air at, say, 102kHz, with its output connected to some kind of acoustic rectifier. If your Q was 20, you could tune in to about a 5100-Hz-wide band. This would be enough to receive and demodulate an ultrasonic AM signal with "telephone quality", i.e. low-pass filtered to about 4kHz.
It's feasible to focus a 102kHz ultrasonic signal in air to a spot about 0.3 centimeter across, or to transmit a low-loss collimated ultrasonic beam of such a frequency that's only a few centimeters across. This could allow substantial-distance ultrasonic AM communication through air despite the way that air attenuates high frequencies (about 1.5 dB/ft at 100kHz, and increasing linearly with frequency from 0.5 dB/ft at 50kHz up to 5 dB/ft at 250kHz). For example, if you started with a ten-square-meter dish or phased-array transmitter transmitting at 120dBa (1 W/m², 10 W total) and focused it on a square-centimeter receiver, you'd get an antenna gain of 50dB. If 40dB was an acceptable listening volume, and your "rectifier" was able to recover -10dB of the original signal, you'd need 50dB at the receiver, which means you could afford 120dB of attenuation along the signal path: 80 feet.
At this distance, your Airy disk radius angle (1.2λ/d) is about 1.2 * 0.003 m / √10 m = 0.001, which at 80 feet gives you an actual radius of 3cm, or 6cm diameter. So you're diffraction-limited by your transmitting antenna rather than scale-limited by the wavelength of the signal. This limits your actual antenna gain to 40dB instead of the 50dB in the previous paragraph, so you'd only be able to actually transmit about 75 feet.
If you could get by with a narrower-band signal, you could use a lower frequency. At 51kHz, where you could transmit three times as far with the same path attenuation (at the cost of less antenna gain, since your Airy disk diameter doubles with the longer waves and triples with the greater distance to 18cm, making it 36 times greater in area, bringing the maximum antenna gain down to about 25dB), your Q=20 receiver could handle a 2.6kHz band. If you were only transmitting speech, you could probably get by with that with a simple hack: frequencies over 2kHz in speech are almost always part of a burst of white noise, such as a sibilant. If you hook up a high-pass filter to the decoded signal and run its output to something nonlinear, you should be able to generate strong harmonics up to a few kHz, which would imperfectly approximate the high-frequency component of the original signal. (I think this is the reason that audio clipping in walkie-talkies improves comprehensibility.)
Being able to transmit comprehensible speech, to a passive receiver with no moving or electronic parts, anywhere in a 225-foot radius (comparable to Wi-Fi) sounds pretty cool, even if not a real improvement over just hollering. The receiver needing to be 36cm across would seem to somewhat blunt that, though, although you could get a proportionally smaller receiver by making a proportionally larger transmitter.
With this level of spatial demultiplexing, however, you might not need frequency division multiplexing at all. Even at 26kHz, which ought to give you a greater transmission distance (450 feet?), your 10m² transmitter can focus down to a 1.5-meter-diameter spot. A larger transmitter could both transmit more energy and focus it on a smaller spot.
What if instead of transmitting the signal through free air, you transmitted it down a string, like a higher-tech version of paper cups connected with string?