[KeithEmo]

All of the theory about the difference being inaudible is based on the claim that we humans will be unable to audible distinguish between the original impulse and that new "sinc-like waveform".
However, oddly, I'm not aware of any studies actually done that prove that to be true.
I'm also not sure that the process of converting the impulse to the longer peak will be perfectly symmetrical in time.
(In order to tell if our short impulse occurred at the beginning of the sample, rather than the end, by looking at the longer waveform, we would have to make some assumptions about symmetry.)

I have read at least one study (it was an AES pub) that tested the audibility of "timing cues".
They seemed to have statistics to show that, when a given test sample was recorded at a 96k sample rate, then reduced to 44k, the positions in the sound stage of some sounds changed.
The authors interpreted this to suggest that we actually do notice timing cues of very short duration.

I'm not sure I have a definite opinion on this - but I am not convinced that it doesn't deserve a little more study before we start making assumptions.

Not following you here, Keith. The theory says you must first bandlimit the signal, so your infinitely thin 'click' will stretch out to a sinc-like waveform and thus be amenable to sampling at the given rate. And if the filter you use to bandlimit is linear phase, then there is no time shifting of the peak.

 

[KeithEmo]

The problem is that not everything CAN necessarily be "thought of as sine waves".... at least not if you don't have an infinitely long sample to work with.

Let's take a look at basic Nyquest theorem......
According to "the general truth", if I have three samples, they uniquely define a single sine wave (this is the justification for being able to reconstruct anything at less than half of the Nyquist frequency perfectly).
However, if you read the fine print, this is true for a continuous sine wave ONLY.
(Clearly, if you draw three dots on a piece of graph paper, I can draw any number of arbitrary waveforms that pass through all three.)
The normal logical response to my argument is that "if we band limit everything properly, then only one of those waveforms will fit properly".
However, again, this assumes sine waves..... after all, by definition, and filter we use will blur the signal in time.
(In other words, we CANNOT constrain my arbitrary waveform to the appropriate range of frequencies to deliver only one unique result without altering it in other ways.)

I should also point out that I'm talking in the theoretical realm here.
I will concede that it's quite possible that no such signal will ever occur in nature.
If that's true, then you might make a fair argument that "nobody will ever hear the difference".
However, from a purely scientific point of view, if a single test sample proves to be audibly changed, then you cannot claim that the difference is absolutely inaudible.

If you look very carefully, you'll find lots of other areas where the theory falls down....
For example, DACs don't actually deliver a sin-C function - they approximate one.
The theory also states that, when we band limit my impulse, the resulting waveform will extend into the infinite future and infinite past (at ever decreasing levels).
Yet, when we build an actual circuit, it never delivers an output that occurs before we input the sample.
Instead, what we get is an approximation; it is only accurate forward and back in time as far as the number of taps in our filter allow us to calculate it.
Our conversion from impulse to sin-C approximation would only be perfectly accurate if we had a filter with an infinite number of taps.... and an infinitely long signal.
(And, if the energy from my single-sample impulse is spread over twenty samples, then you're betting that our ears cannot resolve the difference between a one-sample signal and a 20-sample signal as long as the energy works out the same.)

the 20hz to 20khz is the estimated range for the human species, not world record numbers... many young children can hear above 20khz, geezers almost never get past 15khz. still we stick to 20-20k as a reference. the day I learn that musicians and sound engineers manage the ultrasound content by ear, I'll be willing to bother keeping it as "the artist intended it".
most of us here have highres media and use them when they're good. we're against false arguments pretending audibility that is consistently debunked in controlled tests. and we're against the highres industry because they seem unable to keep themselves from lying about what they're selling. I personally have nothing against fidelity, I can hear it or I can't, I keep it or I convert to lower resolution to save space on my hard drive, those are personal choices. I don't know anybody here who's trying to force the all world to use only AAC.

I don't get the infrared thing. maybe the laser was also emitting visible light? did you fact check that?


as for the impulse, again I don't get it. I can't help but think of an impulse in term of sine waves. and from there, if it contains above half sampling frequencies, then of course we'll lose some. so let's say it's not about the false argument to disguise ultrasounds as something audible, and it's a legit concern about the transient of some instruments attack or whatever. if my own ears aren't applying the equivalent of a low pass on those frequencies any way, how come it's so hard to pass a blind test about highres? that's something I can't wrap my mind around, if something is audible and 44.1khz fails to keep it audibly transparent, why is it so hard to pass a blind test vs highres using musical content? my own conclusion is that we're never hearing more of a transient signal than its content within the audible range. talking about audible stuff while failing to hear them in a blind test, that's a contradiction.

 

[KeithEmo]

1)
I meant to suggest that SOME of the information is lost... as opposed to none being lost.

2)
In fact I can produce purely arbitrary waveforms.
I can write a program that generates random 16 bit numbers and writes them to a file - then interpret that file as a wave file.
My point was that the idea of transforming a complex waveform into a sum of sine waves depends on a signal of reasonable length.
I'll give you a totally absurd example......
I have three consecutive samples, sampled at 44k; they are 127, 275, and 111.
Can you offer me a single sum of sine waves that is UNIQUELY associated with those three samples?
(Note that we are not assuming anything about the samples on either side of those three.)
In order to do a proper unique transform you would need to know something about what happened before and after.
(A drum hit is surely not a perfect impulse; but, equally surely, you will significantly alter the time/energy distribution when you band limit it.)

I think it would be interesting for someone to do a full run of tests.....
Let's see if people can distinguish between 44k and 96k sample rate with short runs or tone bursts of different waveforms.
And let's try it with impulses of different duration, band limited to different rates, and positioned at various points in space.

3)
I'm NOT assuming that you can definitely hear it. (I've only heard of one study - and it seemed inconclusive to me.)
However, I'm not willing to assume that you CANNOT hear it either.

Let me clarify my overall sentiments on this subject......
I am not at all convinced either way.
However, I also believe in things like "margin for error".
It seems foolish to me to insist on limiting the bandwidth of a recording to 22 kHz "because we couldn't hear it if it was better".
When I print photos, I don't carefully limit the resolution to that which I'm sure will be visibly better... I use several steps better resolution "just in case".
And, when I buy a ruler, I don't make sure to buy a ruler that's no more accurate than I need.
Since both storage space and bandwidth are now so cheap, I'd rather use 96k instead of 44k, just on the off chance that 44k is less than absolutely audibly perfect.
(The cost of 24/96k vs 16/44k is really negligible when expressed as $/song.)

If, a decade from now, someone discovers that it takes a 46k sample rate to be totally indistinguishable from the original....
I'd much rather have a bunch of 96k recordings that are "unnecessarily good" instead of a bunch of 44k recordings that are "almost good enough".
I just look at it like cheap insurance.

Here's one test I'd like to see performed.
Start with an impulse - the same in both left and right ears.
Now start advancing the timing on one compared to the other.
We should expect the perceived position of the sound in space to shift.
I'd like to know the minimum shift, in microseconds, before it becomes noticeable.
Then I'd like to see if that changes at different sample rates.
When that impulse is band limited, it will be transformed into a wide shallow peak....
The width of the peak will depend on the sample rate - and so the band limit.
Perhaps human ears are better able to differentiate the timing differences with a particular peak shape or width.

We have to be really careful here:
1) "Lost" is a strong word. You mean absolutely ALL information has been lost? Nothing, even stochastically, can be said about the peak location of this packet of data?
2) "Purely arbitrary waveforms" isn't meaningful. White noise looks pretty arbitrary, after all. Plus, you can generate LOTS (official mathematical term) of signals from adding up even a finite numbers of sine waves.
3) You seem to be assuming that you can hear the phase shifts of HF content even if you can't hear the HF content in isolation, which is quite an assumption.

 

[KeithEmo]

You didn't read carefully enough.....

I didn't say " a drum sample" or "a drumbeat".
I said I'm going to tap my plastic drumstick on the metal edge of the drum....
This should produce a single sharp "tick" - which I would expect to be quite short.
And, yes, if you hear a single click, you do generally get a sense of its location in space.

The problem here is that you don't really know what sort of scale a sample or a drum tap occupies. It's all small, so your mind just lumps it all together. But the duration of a drum tap is several orders of magnitude bigger than the sample- it probably spans a thousand samples. And when it comes to the way that we locate sounds in space, it's not by discerning microscopic fractions of a second like a sample. It involves room acoustics and turning our head to be able to parse directionality. Your understanding of frequencies is the same. How high of a frequency is a drum tap? How high is 25kHz as opposed to 20kHz or 15kHz or 10kHz? What do they sound like? I'll give you a hint... they don't really sound all that different because they're all around the highest octave we can hear- and the difference between 10 and 15 is more than the difference between 20 and 25. Try and define what the numbers represent and you'll understand the physics better.

As a for instance... you're talking about super audible frequencies and the audibility of sound above 20kHz. If I remember correctly, the highest frequency ever recorded as being heard by human ears is somewhere between 22kHz and 23kHz. The frequency range doubles with each octave. So the 23kHz represents 1/13th of an octave. There are seven notes in an octave, so that only represents a half of a note. Humans hear 10 octaves at best. So the difference between hearing 20kHz and hearing 23kHz is almost nothing.

Taking that one step further... The difference between 10kHz and 20kHz is one octave. 1/10th of the sound we can hear. Seven notes at the bleeding edge of human hearing. Only a couple of musical instruments produce sound in that range, and most of it is inaudible due to masking. Controlled testing has shown that the presence of super audible frequencies adds nothing to the perceived sound quality of recorded music. They have also shown that most people don't even really care if all the frequencies above 10kHz are rolled off. Super audible frequencies just don't matter.

 

[KeithEmo]

I'm trying not to take this to the level of the absurd......
The author who wrote the article on which the title of this thread is based made the assertion that "humans can't see the frequencies of light emitted by IR remote controls".
I just wanted to point out that his claim is yet another generalization which is usually, but *not* always, true.

alright so it's borderline visible range and we compensate the lack of cells sensitive to that range with high level energy. so about the same idea as making 20khz audible for me even nowadays if I boost the volume to unsafe levels.

 

[KeithEmo]

I tend to get rather long winded... so I wanted to throw in a short answer here for those who've been following along.

Digital sampling theory says that I must first band limit the signal before sampling it.
However, the theory DOES NOT say that the band limited signal will be audibly identical to the original.

A totally different set of studies tell us, pretty conclusively, that we humans are only able to detect continuous sine waves between 20 Hz and 20 kHz (with a few exceptions).
This strongly suggests that the presence or absence of sine wave components above 20 kHz will be audibly undetectable to humans.

However, specifically combining those two to claim that we cannot hear differences in digital signals above 20 kHz, even with non-sine-wave content, involves an inference.
(The specific inference is that, for purposes of human audibility, we can safely treat any digital signal as "continuous sine wave information".)
I do not believe that this inference has been sufficiently substantiated to be accepted as "true".

Not following you here, Keith. The theory says you must first bandlimit the signal, so your infinitely thin 'click' will stretch out to a sinc-like waveform and thus be amenable to sampling at the given rate. And if the filter you use to bandlimit is linear phase, then there is no time shifting of the peak.

 

[bigshot]

I said I'm going to tap my plastic drumstick on the metal edge of the drum....
This should produce a single sharp "tick" - which I would expect to be quite short.

I read you correctly. That sound you describe would register from attack to decay over hundreds and hundreds of samples. A sample is 1/44,000th of a second. That's 20 times faster than a camera shutter. It's a ridiculously small sliver of time. In the real world, you could never make any sound anywhere close to that fast. A little googling would solve all the questions you've got.

 

[danadam]

I'd much rather have a bunch of 96k recordings that are "unnecessarily good"

@amirm looked into some of them, you can see how "unnecessarily good" they are

I'd like to know the minimum shift, in microseconds, before it becomes noticeable.

20 µs according to some (for 16/44, though personally I don't believe it matters).

 

[KeithEmo]

I would go with "very short", and I agree that you probably wont find many - if any - natural sounds that quick.
However, I have a high speed strobe that delivers 2 microsecond light pulses, and it's trivial to generate a one microsecond electrical pulse.

Incidentally, I took your advice and Googled some test results.....
You might find this one interesting....
http://www.aes.org/e-lib/browse.cfm?elib=18296

It's an AES paper from 2016 titled: A Meta-Analysis of High Resolution Audio Perceptual Evaluation

They concluded that a small but statistically significant number of untrained listeners can in fact distinguish between high-resolution audio and CD quality.
Further, they concluded that the ability to hear the difference "improved significantly with training".
(Everyone should read it - the download is free.)

I read you correctly. That sound you describe would register from attack to decay over hundreds and hundreds of samples. A sample is 1/44,000th of a second. That's 20 times faster than a camera shutter. It's a ridiculously small sliver of time. In the real world, you could never make any sound anywhere close to that fast. A little googling would solve all the questions you've got.

 

[RRod]

I would go with "very short", and I agree that you probably wont find many - if any - natural sounds that quick.
However, I have a high speed strobe that delivers 2 microsecond light pulses, and it's trivial to generate a one microsecond electrical pulse.

Incidentally, I took your advice and Googled some test results.....
You might find this one interesting....
http://www.aes.org/e-lib/browse.cfm?elib=18296

It's an AES paper from 2016 titled: A Meta-Analysis of High Resolution Audio Perceptual Evaluation

They concluded that a small but statistically significant number of untrained listeners can in fact distinguish between high-resolution audio and CD quality.
Further, they concluded that the ability to hear the difference "improved significantly with training".
(Everyone should read it - the download is free.)

So basically we have to debunk this meta-analysis? Done on a bunch of non-replicated studies? If so, there's a thread on another site that gives plenty of analysis.

Literally all anyone has to do to please me, and probably others on this thread, is to give a self-honest DBT assessment of their hi-res listening skills on a real hi-res music source on the system of their choosing. That's it.

 

[bigshot]

I would go with "very short", and I agree that you probably wont find many - if any - natural sounds that quick.
However, I have a high speed strobe that delivers 2 microsecond light pulses, and it's trivial to generate a one microsecond electrical pulse.

What do you plan to listen to on your home stereo? Because you'd never find anything like that in music.

By the way, I've checked out that paper in the past. "Meta analysis" means that it isn't a study... it's someone's interpretation of various studies. None of the studies cited in that meta analysis came to the same conclusion as that particular meta analysis. That paper is pretty much disregarded because it cherry picked and massaged statistics to come to the conclusion they did. If you use normal statistical analysis, it comes out that no one can perceive anything above 20kHz... which is basically what all the studies cited in that paper decided.

It's VERY common for people to decide what they want to believe is true and then look for facts to support their conclusion. The problem is, that isn't good science. It's better to assemble facts without a preconceived bias and then see what they tell you.

 

[KeithEmo]

I'm not suggesting that anyone has to debunk anything.....
Nor am I suggesting that anyone buy anything.....
(And, since I haven't done any sort of carefully planned testing myself, I claim no definite opinion either way.)

However, from the information I've read about the various tests that have been performed, to me many of them seem to have been deeply flawed, or too limited, or badly thought out.
This meta-analysis, and the fact that it reached the exact opposite conclusion many of the individual studies reached, just reinforces my notion that the testing so far has been insufficient to prove the point either way.
We have an assumption - that there are no audible benefits to high-resolution audio.... and I simply don't believe that it's been tested thoroughly enough to believe or disbelieve that it's always correct.

I'll also admit to being somewhat puzzled as to why so many of the tests that were run were so deeply flawed.......
It seems obvious to me that a test which would provide relatively conclusive results could be devised without a lot of effort....
I agree with you that it SHOULD be easy enough to prove either way.

So basically we have to debunk this meta-analysis? Done on a bunch of non-replicated studies? If so, there's a thread on another site that gives plenty of analysis.

Literally all anyone has to do to please me, and probably others on this thread, is to give a self-honest DBT assessment of their hi-res listening skills on a real hi-res music source on the system of their choosing. That's it.

 

[71 dB]

However, the theory DOES NOT say that the band limited signal will be audibly identical to the original.

That's actually a good point. We can't store signals 100 % identical to original. Analog is never identical. Digital is never identical. Nothing will ever be. I accept this fact of life and just enjoy the music. It's identical enough for me, gives me the illusion of being 100 % identical.

 

[KeithEmo]

There is a distinct difference between "an interpretation" and a meta-analysis.
In a proper meta-analysis, you are still required to use standard statistical methods, and so the results are in fact still valid.
In fact, the idea is that, because you have a much larger sample, the results are more valid (of course assuming you don't cheat).
(Since they stated their methods, I'm sure a statistician somewhere can point out any invalid assumptions they made, or rules they violated, when making their calculations.)

Personally, I'm inclined to dislike "statistical significance" as a statement of fact.
As many here, I would much prefer actual facts, derived from the results of actual tests.
And it also seems to me that it shouldn't be that difficult to devise a test that would produce conclusive results.
(But I find it sad that most of the tests I've seen published had glaring flaws..... like using "high-res" samples that weren't authenticated to actually be high-resolution at all.)

Unfortunately, there's also a valid issue with "the inability to prove a negative".
If some number of people can consistently hear a difference - then that proves there is a difference.
However, if any number of people consistently fail to hear a difference, we have no way to tell if no difference exists, or if a difference exists, but our test is insufficient to show it.

What do you plan to listen to on your home stereo? Because you'd never find anything like that in music.

By the way, I've checked out that paper in the past. "Meta analysis" means that it isn't a study... it's someone's interpretation of various studies. None of the studies cited in that meta analysis came to the same conclusion as that particular meta analysis. That paper is pretty much disregarded because it cherry picked and massaged statistics to come to the conclusion they did. If you use normal statistical analysis, it comes out that no one can perceive anything above 20kHz... which is basically what all the studies cited in that paper decided.

It's VERY common for people to decide what they want to believe is true and then look for facts to support their conclusion. The problem is, that isn't good science. It's better to assemble facts without a preconceived bias and then see what they tell you.

 

[bigshot]

Are you sure you aren't cherry picking? It sure seems like it.