for some music, especially classic, 18 or 20 bits would be great.
The quieter parts might need more bit depth if they were on their own, but if the whole piece is normalized, at what volume are you listening that you can hear the quantization on the low volume parts?
If you hear rustling in the bushes behind you sometime in the next few days, it's a classical music recording engineer sneaking up behind you to garotte you with a mic cord because he's taken deep offence at the suggestion that a classic recording engineer would
ever normalize the levels on a recording.
I can't think of an actual piece where this happens, but it's quite plausible that there exists a piece of music that has both a solitary piccolo solo and full orchestra playing
fortissimo [For the non music readers that's the Italian notation for
very loud]. Under those circumstances it's quite plausible that the noise floor becomes audible during the piccolo solo.
Lets see how that pans out in numbers.
You can debate the dynamic range of human hearing, but it's at least 120 dB. The dynamic range of a full orchestra is about 50 dB (again, open to debate). Domestic background noise when it's 'quiet' is usually around 40 dBA SPL. 16-bit SNR is 96 dB, 24-bit SNR is 144 dB.
So 16-bit with a 50 dB orchestral dynamic range would leave that theoretic piccolo 46dB above the recording noise floor, for 24-bit it'd be 94 dB above the noise floor. It's quite clear that 94 dB SNR is a non-starter, you'd never hear the noise. What about 46 dB SNR? If you set up so that the piccolo was clearly audible, let's say 60 dBA SPL (quiet speech) the noise floor of a 16-bit recording would be -26 dB to the noise floor of the listening room and the full orchestra would be at 110 dBA SPL, just 10 dB shy of most people's auditory pain threshold.
Conclusion: 16-bit is adequate for even the widest dynamic range of orchestral music when listened to in domestic surroundings with 40 dBA SPL background noise. Even in a music studio listening room (typical background 20 dBA SPL) it would still be acceptable. If you wish to listen to full dynamic range orchestral music in an anechoic chamber,
then you will need 24-bit encoding.