Ah, okay, so you're saying ultrasonic frequencies can present audible artifacts in the range of hearing, and if you filter them out before sampling, those colorations disappear?
Not exactly, that's linear thinking again. Consider how a frequency like 23.123456Kz is represented when the sampling rate is not an integral multiple. The ear can hear that frequency as a pure tone. That is, in principle there is enough information there to infer that it is a pure tone. But the digitized version doesn't play that tone, it plays a mixture of tones with different phases and amplitudes (the convolution of Sin(x)/x with a pure frequency, and only frequencies that happen to be at integral multiples of the sampling rate. The ear can infer it's not a pure tone.
Beyond some sampling rate the ear won't be able to do that trick. But I don't know what that rate is. It might even be less than 20Khz or it might be more.
The nyquist theorem applies when the frequency basis set that generated the time series of equally spaced intervals, was drawn from equally spaced frequencies separated by 1/T. But it doesn't apply if the basis set includes off latice frequencies.
Now a person thinking linearly will say to resolve those off lattice frequencies is equivalent to hearing ultrasonics. It isn't. What it means is to reproduce those off lattice frequency generated time series with on-lattice points you would need to have more frequency components.
Y'know personally, I doubt there's anything I myself could hear at higher sampling rates. I'm just being a pest because people keep insisting that the nyquist theorem applies and therefore there's no possibility that there isn't something being lost by the sampling at constant intervals. That's not true but it also doesn't mean Mr. Young is right either.