Today I found a very well-researched, well written article on xiph.org, that basically said that 16/44.1k should be enough for everybody.
I was slightly irritated, because I’m a firm advocate of DSD and 24/96k; having heard, and been able to spot the differences many times.
The author, Monty, argued that the human hear is unable to hear frequencies above 20k.
He cited the usual, well known studies on the subject, and was kind enough to include how they arrived at their conclusion.
And here is what he got wrong: The studies, upon close and unmerciful inspection, show only that the human ear is incapable of hearing sinus waves above 20k. Well, I’m not going to argue with that, we’re not bats, spot on!
If you know how a Fourier Transform works than you probably already know where I’m going here, but if not, read on:
Any waveform, says Fourier, can be divided into sine and cosine waves of different frequencies, that combined yield the analysed waveform.
A square wave of 10k would have a 10k sine basic and then a 20k sine on top of that, and a 40k and so on, until it becomes a square wave.
Now if you chop off the frequency band at 20k, it just doesn’t look anything like a square wave, and doesn’t sound like one either.
But square waves are about as common in nature as sine waves, so that isn’t much of an argument in the favor of 96k sampling.
But, … transients: sharp, short spikes of audio, happen very often in nature and in music, particularly in modern music. And here 44.1k adds severe distortion, making placid hills out of sharp spikes, while also adding pre- and post ring to the spike, courtesy of oversampling digital anti-aliasing filters.
Now Monty goes on and cites a study by the AES that revealed candidates were only guessing whether something was 96k or 44k in an ABX blind test.
Why? Hard to tell, but reproduction equipment could be the main culprit here. Cheap amplifiers, headphones and loudspeakers often have very poor time-domain fidelity. Meaning? They smear a short, sharp transient all over the place. Making a better time-domain resolution undetectable.
David Blackmer, the eminent founder of DBX, wrote an excellent article on the subject in 1999. Here is a link.
To sum it all up: We need high-def digital, but we also need to focus more on time-domain fidelity. Here’s were the audio equipment industry is not delivering today, mainly because nobody cares. You get a pair of pretty flat frequency response headphones fo 5$, but headphones with great time-domain reproduction are hard to find and expensive.
For loudspeakers Meyer Sounds HD-1 come to mind, at 7000€ a pair…