Not trying make an argument against discrete time sampling...
I'm just saying the real world is not perfect.
Noise is everywhere and all else being equal the higher the sampling rate the lower the in band noise.
How mush SNR is needed is a whole nother question.
How much? Much, much less than 22 or 23 bits, is the answer. The noise level of the ATMOSPHERE comes into play here. Really.
The noise level of the atmosphere AT THE EARDRUM, 20-20K is on the order of 6dB SPL to 8dB SPL. There is no point in being lower than that. 20 bits above that is 120dB above that, or 126dB, which is above any safe listening level at all. If we want to go to 24 bits (not that I think any convertor out there does that), that's 144+6 = 150dB which is:
1) Well above the linear level for atmosphere transmission
and
2) Above the "instant damage to hearing" level.
So. Really. 20 bits is enough for any sane capture, given that the noise level in a quiet listening room is in the 40dB SPL range usually to start with. There's a thread on this somewhere around here.