can we hear above 20 KHz? might be asking the wrong question

[johncarm]

In asking the question, "What sample rate is needed for digital to be perfect," we often refer to the ear's upper bandwidth. If the Nyquist frequency is greater than the ear's bandwidth, we're good.
 
That might be a bad assumption, however.
 
I've been reading through some sound science posts here and there's an amazingly pervasive assumption in the threads here about how the brain works, which I can't imagine could possibly be true, although I would like to see what psychoacoustics or neurology says about it.
 
Here's the basic idea. Start with two signals A and B, and we would like to see if the brain can tell them apart. I'm not talking about "proving" anything "for all humans" etc., just some basic idea of what we need to know to set up some tests, or make some reasonable guesses.
 
In scenario one, A is some white noise at 0 dBFS. Let's say we have a recording of a sine wave at -10 dBFS, called M. We generate B by mixing A and M.
 
Here, we need psychoacoutical facts regarding the ability to hear smaller signals in the presence of larger ones. I'll call this "masking theory" for short. Both A and M are simple signals by comparison to music, and even if masking experiments were done with sine waves or noise bursts, they would be relevant.
 
In scenario two, A is a choir, M is a marching band at -60 dBFS, and B is the mixture of A and M. Now the signals are more complicated, but they share a feature with scenario one, which is that A and M are highly differentiated. Either one can be heard as a coherent, recognizable sound source. And, when mixed, they will "clash" rather than blend into a single apparent phenomenon.
 
In both scenarios one and two, it makes sense to view signal B as "A plus something added" and to examine the features of the signal that was added. The subtracted signal B-A is something that makes sense to the brain in its own right.
 
In scenario three, A and B are the output of two different amplifiers, both fed with the same input. Now this is quite different, because there simply is no meaning to B-A. The subtraction is hardly relevant to how the ear hears A and B. This means that psychoacoustical research on "small signals in the presence of larger ones" would have to be applied selectively, and only then with a good justification (such as a model of the ear that justifies it). 
 
In scenario four, A is some music that is recorded with a bandwidth of 100 KHz, but then band limited to 20 KHz, and B is the same recording but band limited to 40 KHz. 
 
This is where it's tempting to say "B-A is entirely ultrasonic," therefore the difference can't be audible. But that is conflating two different questions. The first question is "Can I hear a pure sine wave in the range 20 KHz to 40 KHz?" to which the answer is probably "no." The second question is "Can I tell A and B apart when it happens that the difference between them contains only components above 20 KHz?" That is a different question with possibly a different answer. That is the real question to ask about higher bit rates.

 

[Speedskater]

I would go in the opposite direction.
a] A young person may hear a tone above 20 kHz (and maybe well above 20 kHz) in laboratory conditions. The excepts would be young people that have discovered headphone music and those that live in high noise pollution areas.
b] They will not hear those high frequencies when masked by louder lower frequency sounds. And there is a long recovery time from these louder lower frequency sounds.
c] None of this matters in musical situations.

 

[johncarm]

  I would go in the opposite direction.
a] A young person may hear a tone above 20 kHz (and maybe well above 20 kHz) in laboratory conditions. The excepts would be young people that have discovered headphone music and those that live in high noise pollution areas.
b] They will not hear those high frequencies when masked by louder lower frequency sounds. And there is a long recovery time from these louder lower frequency sounds.
c] None of this matters in musical situations.

 
Note that the central point of my post was to ask the right question. I don't know if you meant to answer it, but you did provide an answer. Which is to refer to "masking theory" (I'm not sure if that is the right term, but basically it's the idea that louder sounds mask quieter sounds).  
 
What I have read about masking theory is that it starts by decomposing a signal into its spectral content; that is, it's an operation that's based on a frequency-domain analysis of a signal. Is that true?

 

[Joe Bloggs]

Masking theory says your scenario three is harder to tell A and B apart in than scenario one or two, even when the error signal is of the same magnitude. What has driven you to think the opposite?

 

[johncarm]

Masking theory says your scenario three is harder to tell A and B apart in than scenario one or two, even when the error signal is of the same magnitude. What has driven you to think the opposite

I was giving the example to try to focus on the big picture, the forest and not the trees. Let me try again.
 
We have two devices, A & B. We are asking the question "Can they be distinguished by sound alone?" We then measure A, B, and A-B.
 
Why are we asking this? Why are we doing this?  Stay focused on the practical reality. The answer is that there could be several reasons we are doing this. We could be sound scientists developing theories about the ear and brain. We could be applied scientists developing theories about the behavior of electronic devices. We could be engineers attempting to design a device, and we are measuring/testing it to check how it actually behaves. We could be executives at a consumer electronics company deciding how to market the device.
 
Why would we measure A - B? Maybe we are going to try to answer the question about sonic differences by theory alone without running a listening test. That might be good enough in some cases. Or maybe we are preparing to run a listening test and we need to know what test music to select, so we start by characterizing some of the differences between A and B.
 
Let's say you are a scientist, engineer, or executive in one of these scenarios. I send you a plot of the signal A - B. Nothing else. Does this help you in any of these scenarios?

 

[spruce music]

I was giving the example to try to focus on the big picture, the forest and not the trees. Let me try again.

We have two devices, A & B. We are asking the question "Can they be distinguished by sound alone?" We then measure A, B, and A-B.

Why are we asking this? Why are we doing this?  Stay focused on the practical reality. The answer is that there could be several reasons we are doing this. We could be sound scientists developing theories about the ear and brain. We could be applied scientists developing theories about the behavior of electronic devices. We could be engineers attempting to design a device, and we are measuring/testing it to check how it actually behaves. We could be executives at a consumer electronics company deciding how to market the device.

Why would we measure A - B? Maybe we are going to try to answer the question about sonic differences by theory alone without running a listening test. That might be good enough in some cases. Or maybe we are preparing to run a listening test and we need to know what test music to select, so we start by characterizing some of the differences between A and B.

Let's say you are a scientist, engineer, or executive in one of these scenarios. I send you a plot of the signal A - B. Nothing else. Does this help you in any of these scenarios?

If I know either A or B along with A minus B I can tell you with high reliability if they are audibly different or the same.

So let's say A is a $10k amp while my R&d guys say they can make a $2 k amp that has a difference signal of -90 dB compared with A. That would mean we can sell a $2k amp that sounds audibly identical to the $10k amp.

Oh before Gish galloping further go search for and read about the Carver challenge.

 

[johncarm]

If I know either A or B along with A minus B I can tell you with high reliability if they are audibly different or the same.

So let's say A is a $10k amp while my R&d guys say they can make a $2 k amp that has a difference signal of -90 dB compared with A. That would mean we can sell a $2k amp that sounds audibly identical to the $10k amp.

Oh before Gish galloping further go search for and read about the Carver challenge.

 
So you are saying no matter what A and B are, you don't need to run a listening test to see if they are audibly different?

 

[Joe Bloggs]

It's been demonstrated in many cases of A and B signals that some people are interested in ABXing, that the difference signal, A minus B, is inaudible when played at the same volume that one would listen to A and B at. (same volume in the sense of maintaining the volume relationship between the volume of signal A and the volume of the error signal A-B within it.) Inaudible means that when you turn the difference signal on and off, the subject cannot tell that you're playing anything at all, at any point in time, even in an anechoic chamber. Many people regard such cases of A and B to be categorically impossible to ABX--if somebody shows a positive ABX result for such an A-B pair, it must be because they have done something wrong...

... and the reason for that would be, that people assume that listening for something (a flaw, say, an injected 50Hz hum, similar to the case you put forward) in total silence would be inordinately easier than picking it out from within loud music. Do you think this is not the case?

It's true that in many cases the difference between A and B is not heard as A minus B. It is usually a lot less obvious. Which is the premise for most of the lossy audio compression codecs we have today (what kind of error signals would be least obviously heard, or not heard at all, in the given piece of music / audio “A” to be encoded, and how to make use of this, given that we can encode various pieces of audio "B" that are different from "A" in various ways but would take up much much less storage space in the encoding / decoding protocol we have chosen?)

Blind testing was performed extensively in the research for these modern lossy compression codecs.

 

[spruce music]

So you are saying no matter what A and B are, you don't need to run a listening test to see if they are audibly different?

That's very slightly simplified, but yes. No listening test needed.

 

[johncarm]

That's very slightly simplified, but yes. No listening test needed.

 
 
So would it be correct to say you begin by performing some kind of analysis of A and B? Can you point me to the theory or theories that you would use to answer the question "Are A and B audibly different?"

 

[johncarm]

... and the reason for that would be, that people assume that listening for something (a flaw, say, an injected 50Hz hum, similar to the case you put forward) in total silence would be inordinately easier than picking it out from within loud music. Do you think this is not the case?

It's true that in many cases the difference between A and B is not heard as A minus B. It is usually a lot less obvious. 

 
It seems like a reasonable first guess that an injected signal is easier to hear in total silence, but I would not assume that to be the case in all situations without some kind of guiding theory, confirmed by testing. 
 
What theory would you invoke?

 

[Joe Bloggs]

... and the reason for that would be, that people assume that listening for something (a flaw, say, an injected 50Hz hum, similar to the case you put forward) in total silence would be inordinately easier than picking it out from within loud music. Do you think this is not the case?


It's true that in many cases the difference between A and B is not heard as A minus B. It is usually a lot less obvious. 

It seems like a reasonable first guess that an injected signal is easier to hear in total silence, but I would not assume that to be the case in all situations without some kind of guiding theory, confirmed by testing. 

What theory would you invoke?

I would suggest that you search the literature for any sign of "inverse masking", or any sign that any scientist has been able to find a masking signal (in your case music) that *increases* rather than *decreases* the ability of one to detect a target audio stimulus (in your case the error signal, A minus B).

When the "engineer" of a major audio company treaded on my thread with outlandish claims, including claims that one could hear distortion in music even better than the distortion by itself, I called him up on it, asking him to show any literature that showed this to even be possible. He has remained notably silent.
http://www.head-fi.org/t/769647/objectivists-board-room/1635#post_12203380

 

[johncarm]

I would suggest that you search the literature for any sign of "inverse masking", or any sign that any scientist has been able to find a masking signal (in your case music) that *increases* rather than *decreases* the ability of one to detect a target audio stimulus (in your case the error signal, A minus B).

When the "engineer" of a major audio company treaded on my thread with outlandish claims, including claims that one could hear distortion in music even better than the distortion by itself, I called him up on it, asking him to show any literature that showed this to even be possible. He has remained notably silent.
http://www.head-fi.org/t/769647/objectivists-board-room/1635#post_12203380

But clearly you think the existence of an "inverse masking" signal is unlikely. What theory are you referencing when you make this conclusion?

 

[Joe Bloggs]

But clearly you think the existence of an "inverse masking" signal is unlikely. What theory are you referencing when you make this conclusion?

Not theory, empirical results. Decades of auditory experiments have only revealed masking, (on which masking theory is built, not the other way around) no inverse masking ever found.

 

[johncarm]

Not theory, empirical results. Decades of auditory experiments have only revealed masking, (on which masking theory is built, not the other way around) no inverse masking ever found.

Okay, good.
 
I know I sound pedantic but the problem is that on this forum, I get replies that try to pull me off my central question. The only thing that seems to work is being methodical. 
 
Of course you can say anything you want. You can answer or not answer. I'm just trying to make my central question clear.
 
I know you are asking what the heck my question is. I'm taking a big picture view of audio theory, design, and testing. Notice that audiophiles and "objectivists" have two different paradigms. They start from different places and end up in different conclusions. So I'm going back to basic questions. Like, "How do we get started answering the question if A and B are different?" How do we test a DAC to see how well it performs?
 
You have made a claim that does indeed help us with these tasks, should it be true. That is, there is little likelihood of any inverse masking effect.
 
That does indeed go to the heart of audio design and testing.
 
But when you say "no experiments have revealed inverse masking," you are expressing a confidence that the area has been explored well. 
 
And that may be the case.
 
But there is a pretty big world of signals, if we consider all possible signals A and B. And there is a big world of listening contexts if we consider all possible listening protocols. 
 
So there is some reason you are confident that this territory has been explored well. Right?
 
I can think of one answer. Linear systems theory tells us that a signal can be transformed into the frequency domain. We know that the ear operates on the frequency domain side of things. Therefore we can be systematic in constructing test signals by dividing the spectrum into bands and providing variations on the amplitude within each band.
 
In other words, if we have some theory about the ear and brain, it helps us to be systematic in exploring the likely territory. Agree?