Oddly enough, BBC Radio 4's daily consumer affairs programme actually did a blind test between 16bit 44.1K CD and 24bit 92KHz "HdCD" within the last couple of weeks. Presenter plus a couple of advocates of high-definition CD. With an invitation to the listening audience to see if they could tell the difference (the broadcast signal is heavily compressed and broadcast as medium bitrate DAB or long wave so not really very hi-fi...).

Very interesting test. The clincher for me was when they used a classical track which one of the HD advocates said he'd actually recorded himself so it would be embarrassing if he got it wrong.

The track was played, once through for each bit/sample rate. The presenter and both HD advocates all agreed that the first time through sounded better. HD guys waffled on about the clearly better dynamic transients, fuller-sounding upper harmonics etc.

To be told the track they had all picked was actually the 16/44.1 version.

At which point one guy said he'd been fooled by the Fletcher-Munson effect and the other commented that the studio monitoring supplied by the BBC (who are perhaps the world standard in broadcast radio engineering) simply wasn't good enough to do the HD audio justice.

Here are the facts:
 
1. If you know in advance that a signal is a sine wave, you only need 3 samples to know everything about that sine wave - the frequency, the amplitude and the phase at which each sample was taken. It doesn't matter if it's a low frequency with thousands of samples per cycle or a high frequency with only 3 samples per cycle and it doesn't matter where the samples are taken (i.e. you don't need the samples to be at a zero crossing or at a 90 degree peak or whatever). 
 
2. A sampled waveform is not reconstructed by drawing lines between adjacent samples despite what many think.
 
If a sampled signal contains no frequencies greater than half the sample rate, everything between the samples is in fact stored in the samples. That's the essence of the sampling theorem. 
 
Note that the stuff between the samples is not just stored in the adjacent samples, but in a long stream of samples. If you can't fathom how that could possibly work, lets just say it involves non-trivial math.

I have seen several tests like this so this does not surprise me. One test was by a high-end audio mag (SOS?). Top of the line playback equipment.  World class source material. Music production people, magazine audio critics, musicians,  and members of the public were asked to pick the hi-sample rate, hi-bit depth material vs the 44.1Khz/16 bit material. The results were completely random. In this case, the author attributed the random results to room acoustics...comb filtering. I would probably buy that explanation over the Fletcher-Munsion one. The fact of the matter is we as a species can rarely hear anything over 20Khz and under -96db.
Of course if this was commonly accepted there would be no market for those 'designer' R2/R DACs with the separate power supplies for each bit selling for $20K+.
Michael
 

Which is precisely why this is an endless debate....
 
This debate has existed since the mammals rose to power, and will be continued by the cockroaches that survive our collective passing of the planet to them.
 
People think what they wish, and in all of the thousands and thousands of posts on this subject I have seen, not ONCE did I see a single person ever change their mind, one way or the other.
 
There are multiple positions on the mathematics, on the range of human hearing, and actually any other aspect of this that anyone has floated.
 
It is as it is, and will always be what it has been and never was.
 
All of it is inaudible to me anyway, having grown up with a close and scratch phonograph.
 
Bob Bone
 

drew,
There ARE lines being drawn between the samples. The DAC is interpolating with typically an oversampled sinx/x function and/or a lowpass filter is smoothing the output. The lines between the samples are just not straight lines.
 
Michael

microapp
drew,
There ARE lines being drawn between the samples. The DAC is interpolating with typically an oversampled sinx/x function and/or a lowpass filter is smoothing the output. The lines between the samples are just not straight lines.
 
Michael

Yes. I was attempting to give a less technical explanation and talking about the common perception that the between the samples stuff is reconstructed by linear interpolation.
 
And we should also note the values between the samples can be greater or less than the adjacent samples.

This is an interesting article along these lines:
 
http://mixonline.com/recording/mixing/audio_emperors_new_sampling/
 
I also performed a very similar test. I have a very high quality turntable with expensive pickup cartridge, arm and RIAA pre amp Equaliser. I fed this to one side of a switch box. I also ran that signal through A to D and D to A at 16 bit 44.1 Khz and fed that signal to the other side of a switch box.
 
I blind switched in a room full of very good people with good ears. Very fine monitors in very nice acoustic environment. Switching was seamless and levels were all perfectly matched of course.
 
I use Sheffield Lab vinyl. (possibly highest quailty there is, direct from the studio to the cutting lathe, no tape in between! Unbelievable quality, you have to hear it to believe it!)
 
Very few could pick the digital path. Most had no idea what they were listening to. I have also used high quality reel to reel masters as the source and the same thing happened. The analog signal represents the finest signal there is really.
 
Moral of the story is stop wasting time worrying about digital sampling rates and get down to the music. It is much more important. (and all the other production stuff in between that has a huge effect on the outcome) Bit depth is way more important. 24 bit is better than 16 bit that is for sure. (means lower digital levels and you can use the K system properly) Higher sampling rates are questionable. Some experts say all we need to 50K to 60KHz and that is it.
 
Some converter designs sound better at 44.1 than they do at 96K and it is also the other way around. It is difficult to make a converter sound great at all sampling rates.
 

And the short, non-technical explanation the impact of higher/lower bit depth and sample rate for playback (not talking about processing inside a DAW here, which is more complicated):
 
Assuming dither is properly applied, more bits just equals less noise.
 
Higher sample rates equals higher frequencies can be present and potentially any artifacts near the top of the frequency range are moved to a higher frequency, and also potentially less audible noise.
 
 
The key question in all of this is at what point of these things are any improvements in noise/artifacts/higher_frequencies no longer audible.

I agree, there is no real point for higher sampling rates/bit depths in the final audio. In certain cases there is a point for recording/mixing. Sony/Philips did their homework when they decided on 44.1Khz/16-bits. Claude Shannon laid it all out in the 30's.
 
Michael