Quote:
...The only thing a higher bitrate will give you is a higher upper frequency limit ...
Well not quite true. The Nyquist theorem does not imply this; some posters here do not understand the theorem fully I'm afraid.
The theorem is an exact result in a theoretical world that is only an approximation to reality. You see, the Fourier transform of a bandwidth-limited function has infinite support in the time domain. So unless you believe the song you are listening to began at the dawn of time, and will go on forever, then Nyquist does not apply precisely.
Read
http://en.wikipedia.org/wiki/Nyquist%E2%80%93Shannon_sampling_theorem#Practical_considerations
It's a really good model of nature, so Nyquist packs a lot of practical wisdom ... but no guarantee that you will get an exact reconstruction from 0 to 22.05 if you sample at 44.1 . You won't.
There's more to say: if the original recording A-to-D step is done at 96, which it often is, you will get a better reconstruction of the analog waveform in the D-to-A step if you keep the 96 digital sample. Down converting to 44.1 will lose information that results in a less-perfect analog output after D-to-A.
The difference may or may not be audible. I make no claims about that. But computer audio makes it trivial to store and use music at the same sampling rate it was recorded at, so this is where the world should go, and is: the future is 96 downloads.
I am sure the difference between 16 bits and 24 bits is not audible, but that's not the point. The point is: don't convert. Every conversion loses something, making the final D-to-A less accurate. Again, it may not be audible, but if recording hardware runs at 96/24, then why not run your playback software like that.
Real-world D-to-A conversion algorithms typically begin with an interpolation step that upsamples dramatically, so that the final modulation resulting in the analog signal is more accurate. (Not all DACs do this, but most do, and the ones that don't typically sound worse, most people agree -- some disagree, but that's life). This is the problem with having both the 44.1 / 88.2 / 176.4 world vs the 48 / 96 / 192 world. Upsampling (and downsampling for storage considerations if you have devices with limited disk) across those two families introduces a lot more error than within those families -- interpolation vs simple decimation (really hexamation) and replication.
Half-sampling a 96 file to store as Apple Lossless (48) sounds better to me than the same track off a redbook CD at 44.1 re-sampled by SOX to Apple Lossless at 48. So maybe the difference in sampling rates is audible, but the test was not blind so I don't know for sure. Often playing directly the 96 on a computer (instead of 48 on an iPod) doesn't sound different that playing the 44.1 on the same hardware, so who knows. My belief is the non-congruent conversion is a bigger factor than the increased sampling rate. BTW, I never hear any difference when I convert 24 bit 96 to 16 bit 96.
Back to the main point. In the real physical world, a higher sampling rate, if preserved throughout the playback chain right up to the final D-to-A, can indeed lead to more accurate reconstruction of the original analog signal, and NOT just at higher frequencies. This does not violate the Nyquist theorem. There is no guarantee of exact reconstruction (in the real world) of frequencies at or below half the sampling rate, although you can get very close. I have no idea if faster samplng makes an audible difference. Most published blind tests suggests it does not. So do my own non-blind tests on myself. But my own non-blind tests on myself suggest that sample rate conversion does introduce (small, and only sometimes) audible effects, so that 96 recordings should be kept at 96.
