Glad to know I'm not the only guy who's nerdy enough to spend freetime playing around with a converter (dBpoweramp in my case), converting things down to 8-bit and then seeing what I can and what I can't hear a difference on
So seriously though. 24-bit and 32-bit. Over 140 and 190 dB of dynamic-range, REALLY? As you said, fully-utilizing that wouldn't just be stupid, it would be physically dangerous, and in the case of the over 190dB range for 32-bit, actually DEADLY. As in, sound on that level would literally be the same thing as (the sound from) shockwave from a fricking nuke going off in your face, it would reduce human flesh, blood, and bone to an indeterminable pile of mush, lmao. It wouldn't be the same as experiencing the shockwave itself, but would be like the sound from it, and the effect ont he human body would pretty much be the same, i.e. it'd get ripped to shreds.
http://www.gcaudio.com/resources/howtos/loudness.html
"140dB--Loudest sound recommended for exposure WITH hearing-protection."
"180dB--Death of hearing tissue."
"194dB--LOUDEST SOUND POSSIBLE!"
Loudest sound possible means just that. . .sound at that level would create waves with such a large amplitude that their troughs would become actual VACUUM, in the atmosphere at average pressure at sea-level here on earth. Any "sound" beyond 194dB in our atmosphere at sea-level would no longer actually be a soundwave, but rather an actual shockwave, in terms of the physics of what would be happening. So that yes, sound in the range of 190 to 194dB would have the same general effect on living tissue as the shockwave from a large explosion. Meaning that 32-bits of dynamic range could literally reduce a human being to nothing more than little teeny bits and pieces. Lol.
You're right - there is almost certainly no need for a dynamic range even as wide as that offered by 24 bits of depth in a recording that you're simply going to play as is. However, there are two "exceptions". First off, you need to differentiate between overall dynamic range and "local" dynamic range" (meaning dynamic range over a short time). In the case of listening to music, what that means is that we've all been assuming that you're going to turn the music on and let it play - so you're going set the level to where the loudest parts are comfortable (or, at least, non-lethal). However, not everyone listens to music that way all the time. So, for example, if you were to start listening to your favorite classical CD, with the level set so that the loudest parts were quite loud, the 16 bit CD audio would have plenty of dynamic range. But, if you manually turn it up 30 dB on the quietest spots "to hear them better", then the noise floor, or even digital artifacts, may be audible at that point - because, by raising the gain by 30 dB, you've also raised the noise floor by 30 dB (which makes the "effective S/N ratio" of your CD at that moment about 65 dB). So, if you're the kind of person who "turns it up to hear the details in the quiet spots", then using 24 bits, or even 32 bits, will prevent this from happening.
The other case is really the same thing - but when you're making a recording or editing it. If you've ever made your own recordings, then you know that you can't always predict exactly how loud things will be in advance. In a real studio setting, with time for sound checks, this isn't usually a huge problem. However, in less formal settings, it's usually very difficult to predict what's going to happen, and not at all uncommon to find yourself diving for a level control when things get loud (assuming you prefer to avoid limiters and compressors). Room acoustics change - a lot - when the audience arrives, and suddenly finding yourself recording from the fifth row instead of the twentieth can make a huge difference. You may also find that things get a lot louder when the commentator stops talking and the music starts.
If you've ever tried to record using a cassette deck, or a digital recorder that only supports 16 bit audio, then you know how difficult it can be to select a recording level low enough that you're positive you won't get any overload, yet high enough to avoid the noise floor of your recording equipment. Even though most micrphone preamps don't really have 96 dB of dynamic range anyway, between the various level controls on the mic preamp, the input pads on your equipment, and the unknowns of your source, having a recorder that operates at 24 bits makes it a lot easier to pick a level that's safely below even the remote possibility of overload, yet safely above the noise floor. (This is partly because you have more adjustment range to use, and partly because, even though portable recorders really don't have much more than 90 dB of "real" dynamic range, ones designated as 24 bit devices usually are quieter than 16 bit ones.)
The same is a lot more true during editing - where different tracks may be boosted or cut in level - sometimes drastically. With editing, you also have the issue that any noise or artifacts that are present may be "magnified" when you add tracks together, and many types of processing actually boost the audibility of certain types of artifacts, as well as introduce new artifacts and mathematical rounding errors of their own. If you're mixing multiple tracks, recorded at different levels, and applying different processing to each, it's not unusual to raise and lower the level many times, and over a very wide range, over the entire process. And it's pretty obvious what's going to happen to the noise floor if you take a drum recording, recorded at a relatively low level to begin with, and boost the treble by 15 dB to "sharpen it up a bit". (I'm sure you've heard otherwise good sounding commercial recordings where the background noise jumped in level when a certain instrument or performer joined the mix; and various types of dynamic processing are well know to introduce "breathing" if the noise floor is even slightly audible before you apply them.) Making sure that the noise floor is
FAR below audibility provides a safety margin against this. This is why most audio editing software actually converts the audio to be edited to 32 bits, or even 64 bits, and operates on it at that bit depth.
This isn't entirely limited to "process editing" either. For example, if you play a two-channel recording in simulated surround using Dolby PLIIx, you may find that certain artifacts and types of noise are exaggerated - because the decoding process "pulls" those sounds to the rear channels and raises their level. This was a major problem in the latter days of SQ4 four channel - where the typical decoder would specifically "misidentify" background noise and "place" it in the rear channels at boosted and varying levels - the result was that what sounded like relatively innocuous tape hiss or record surface noise in the front channels might end up sounding like a very audible and annoying desert sand storm in the rear channels. And, in modern equipment, you may find that your room correction system has "eaten up" 10 or 12 dB of dynamic range by boosting the high frequencies to compensate by a slight high frequency droop in your speakers, or a little extra absorption in your listening room.