[ EDIT: contents of this post were factually incorrect. What Kazinator wrote in response is correct, and I don't want to leave incorrect ideas floating around. ]
If the goal is audio capture and reproduction, the filter used when sampling at 96 kHz still starts rolling off at around 20 kHz. The wrap-around aliasing artifacts do not begin until around half of 96 kHz, or around 48 kHz. So the filter only has to be steep enough to hit a large attenuation at 48 kHz: full signal to virtually nothing over the space of 28 kHz.
If you're sampling at 48 kHz, the signal has to be severely attenuated already; it has to go from 20 kHz to deep cutoff in just the space of a few kHz.
At 96 kHz can achieve the effect as if you were sampling at 48 kHz, with a steep filter. You sample at 96 kHz with a milder filter, and then purely in the digital realm, you down-sample to 48 kHz. There is an overall filter consisting of the original analog one plus the digital processing.
That is cheaper and more reliable than doing it all in analog.
An analog filter with a steep cut off will be challenging in mass production because of the strict component tolerances.
Sure, you could use a steep filter with 96 kHz also. Say, a steep filter that starts cutting off at 30 kHz. It would still be a less demanding filtering application because of the margin that you have in the frequency domain. The multiple poles of the filter don't have to be lined up as well. E.g. if the first pole starts rolling off at around 30 kHz, and then next ones at 31, and the third one at 28, ... it doesn't matter because you're still hitting the absolute target of there being next to nothing at 48 kHz, and nearly the full signal at 20 kHz.
Sure, but going to disagree about the numbers there. I have a DAW interface which is exceedingly linear (+/- <0.15 dB) up to 35 kHz. Don't need to start filtering so low to hit those results.
This explanation loses me at "infinity per dB" ... and I'm an EE. I think you're trying to cover too much ground here, it's really confusing to try to understand what you mean.
I believe the comment you're responding to is talking about the analog filter that is needed to avoid aliasing -- as the first words of your comment correctly note/explain.
And in particular, the original comment seems to be noting the phase distortion (in frequencies near the cutoff) that analog brick-wall filters will cause. This has been a design contention for decades, really, ever since the CD format was introduced.
It's a big design space, with options for gentler analog filters, followed by very fast digital sampling, and further tricks with filtering in the digital domain, where you don't have to worry about getting great capacitors, etc.
It may be out-of-scope to lay all that out in one paragraph!
You're right that its a big design space. The key takeaway is that "yes, higher sample rates can actually make a difference, but almost entirely down to the filter design, not because Nyquist moves ... and you probably cannot hear the difference."
