Blue or red pill?

[PeterSt]

6 minutes ago, sandyk said:

Some will even make that  claim for -130dB of difference.

 

I just tried to mention something inordinate.

But you got it, Alex.

[manisandher]

8 hours ago, fas42 said:

... and conversely with noticeably audibly distinct samples, show extremely close matching by the program, with a tiny difference which has the crucial difference mixed in with benign aspects!

 

From what I've seen on various forums, this is where the 'experts' jump in and say that the "noticeably audibly distinct samples" must have been imagined by the listener, because the diff program shows such a small difference between the samples that can't possibly be heard. I think we all agree that -160dB, -130dB, etc., differences cannot be heard.

 

But these 'experts' also swear by properly-conducted A/B/X tests. So they're not going to be able to dismiss our results as my having simply imagined things.

 

The issue now is that no matter how accurately a diff program can find differences, it will likely not find consistent differences between the analogue outputs of the two playback means we used. That is, any differences it finds between 'A' and 'B' will likely be exactly the same sort of magnitude as the differences it finds within all the 'A's and within all the 'B's. (I suppose the differences will be > -96dB, owing to the 16-bit DAC used.) In which case, any such diff program, no matter how accurate, would totally fail the A/B/X test.

 

The only conclusion would be that it's measuring the wrong thing.

 

Mani.

[adamdea]

20 minutes ago, manisandher said:

 

From what I've seen on various forums, this is where the 'experts' jump in and say that the "noticeably audibly distinct samples" must have been imagined by the listener, because the diff program shows such a small difference between the samples that can't possibly be heard. I think we all agree that -160dB, -130dB, etc., differences cannot be heard.

 

But these 'experts' also swear by properly-conducted A/B/X tests. So they're not going to be able to dismiss our results as my having simply imagined things.

 

The issue now is that no matter how accurately a diff program can find differences, it will likely not find consistent differences between the analogue outputs of the two playback means we used. That is, any differences it finds between 'A' and 'B' will likely be exactly the same sort of magnitude as the differences it finds within all the 'A's and within all the 'B's. (I suppose the differences will be > -96dB, owing to the 16-bit DAC used.) In which case, any such diff program, no matter how accurate, would totally fail the A/B/X test.

 

The only conclusion would be that it's measuring the wrong thing.

 

Mani.

Any statement that “the only conclusion is” is almost certainly false. 

[manisandher]

5 minutes ago, adamdea said:

Any statement that “the only conclusion is” is almost certainly false. 

 

I said "would be", not "is".

 

Anyway...

 

All men are mortal.

I am a man.

The only conclusion is that I'm mortal.

 

"The only conclusion is" is perfectly valid in  'deductive' reasoning. But I agree, not so in 'inductive' reasoning.

 

Mani.

[adamdea]

I see no relevant syllogisms. 

[manisandher]

1 minute ago, adamdea said:

I see no relevant syllogisms. 

 

The signals were bit-identical, but audibly different

The measurements show no differences

Therefore...

 

Mani.

[esldude]

3 minutes ago, manisandher said:

 

The signals were bit-identical, but audibly different

The measurements show no differences

Therefore...

 

Mani.

Do we know the measurements show no differences at this point?  By that I mean something other than run to run differences at very low levels that will be due to analog thermal noise and such.  

[manisandher]

7 minutes ago, esldude said:

Do we know the measurements show no differences at this point?

 

I was just putting the syllogism forward to Adam. The final premise is still to be demonstrated. That's why I said:

 

57 minutes ago, manisandher said:

The only conclusion would be that it's measuring the wrong thing.

 

Mani.

[PeterSt]

11 minutes ago, esldude said:

By that I mean something other than run to run differences at very low levels that will be due to analog thermal noise and such.

 

And that is exactly the problem. Those differences ARE there as a fact and you can't do a thing with it for that reason. This is apart from the impossibility to align two different takes (only infinite ADC sampling rate would allow for that (so Mani, now you know).

[manisandher]

5 minutes ago, PeterSt said:

Those differences ARE there as a fact and you can't do a thing with it for that reason.

 

But the exact same differences were coming through my ears too, right? (I mean due to the DAC, not the ADC.)

 

Mani.

[PeterSt]

1 hour ago, manisandher said:

the diff program shows such a small difference between the samples that can't possibly be heard. I think we all agree that -160dB, -130dB, etc., differences cannot be heard.

 

It is not even about that. Well, see the emphasis; you feel it is about that part, but the further interpretation of it is a bit difficult. Look :

 

If two adjacent samples show a difference (for two runs I mean) then no SPL is going to catch that. I say "SPL" on purpose because that is what we should perceive audibly and you already can't imagine, while something like a diffmaker program works with dB's which are not "detectable" as such. Thus, when something which should be 800uV turns out to be 860uV (which is a most normal thing to happen) then no dB figure will show you this. It can be calculated easily though, from theory. Keep in mind that we talk about one sample to sample difference here.

When this one sample to sample difference starts to have more real impact because it happens one out of 10 samples and is more realistically to happen (you will believe in the reality of it) it still will be so that no diffmaker will tell you that the now more average dB difference of 40uV/10 (!) is a difference to begin with.

Still it is though, because it is about such differences we do perceive. Remember (Mani), I did all this - so I exactly know what I am talking about plus I know what it *is* about. I showed that plot for the 25th time or so for a reason. Don't look at it as a nice plot without plot. It shows the plot. There the differences readily show. Not in dB but in samples ...

 

Let's for fun think about how dither works (out) ...

[PeterSt]

5 minutes ago, PeterSt said:

Let's for fun think about how dither works (out) ...

 

I make this up on the spot, so ready to receive some bullets ...

 

If we have a general noise level of say -120dB (distance to max output, so we're talking about a SNR thing), then the (24 bit) signal which also plays under that noise level can be made visible because an FFT analyzer can do that. Our ears can do this too because we already accept that dither helps us (out ears) to hear through the noise (ever back REM offered a test file for this, and probably such tests still exist).

 

So can we agree that we can perceive music which is even under the general noise level ? Mind you, of course the music is not noise as such and it just *is* there in differentiation from the noise (the FFT is perfect in showing that - without me explaining currently how but you will believe me).

And how do you think "diffmaker" works ?

It will fail all over.

 

Does this help ?

 

 

[PeterSt]

2 minutes ago, PeterSt said:

It will fail all over.

 

We see noise only. We compare noise only. The guys of fame will tell you that you won't be able to hear differences at that level. Whether 160dB or 120dB. Noise is noise. But it isn't.

[manisandher]

9 minutes ago, PeterSt said:

... see the emphasis [small difference between samples]...

 

Peter, when I used "samples" here, I was using Frank's terminology. I didn't mean 'samples' in a digital sense, I meant samples in an A/B/X sense, i.e. random samples. Mans and I conducted 10 random samples of X in our listening tests. So when I said "small differences between samples", I meant small differences between 'A' and 'B'.

 

Sorry, I should have been clearer.

 

Mani.

[Summit]

I take the green pill!

[PeterSt]

19 minutes ago, manisandher said:

Peter, when I used "samples" here, I was using Frank's terminology. I didn't mean 'samples' in a digital sense,

 

Yes. But sadly I do.

Now read my post(s) again. 

[manisandher]

2 minutes ago, PeterSt said:

Now read my post(s) again. 

 

Oh I already did... many times.

 

I'll save you the hassle:

 

 

Mani.

[PeterSt]

9 minutes ago, manisandher said:

Oh I already did... many times.

 

Haha, I didn't mean that one per se. I referred to those from today - the ones about the sample to sample difference ... OK ?

[manisandher]

15 minutes ago, PeterSt said:

Haha, I didn't mean that one per se. I referred to those from today - the ones about the sample to sample difference ... OK ?

 

Yeah I know.

 

1 hour ago, PeterSt said:

If two adjacent samples show a difference (for two runs I mean) then no SPL is going to catch that. I say "SPL" on purpose because that is what we should perceive audibly and you already can't imagine, while something like a diffmaker program works with dB's which are not "detectable" as such. Thus, when something which should be 800uV turns out to be 860uV (which is a most normal thing to happen) then no dB figure will show you this. It can be calculated easily though, from theory. Keep in mind that we talk about one sample to sample difference here.

When this one sample to sample difference starts to have more real impact because it happens one out of 10 samples and is more realistically to happen (you will believe in the reality of it) it still will be so that no diffmaker will tell you that the now more average dB difference of 40uV/10 (!) is a difference to begin with.

Still it is though, because it is about such differences we do perceive. Remember (Mani), I did all this - so I exactly know what I am talking about plus I know what it *is* about. I showed that plot for the 25th time or so for a reason. Don't look at it as a nice plot without plot. It shows the plot. There the differences readily show. Not in dB but in samples ...

 

So I saved you having to show the plot for the 26th time 

 

Mani.

[fas42]

7 hours ago, PeterSt said:

 

Frank, come on now.

 

Tsk, Peter, I couched my post to indicate that the program could possibly be useful - a person "skilled in the art" may get significant results - I lost interest in using it because it had too many issues.

 

7 hours ago, PeterSt said:

 

It's the other way around. People of perceived fame use it to show you that any -160dB of difference can't be audible and thus is no difference.

In the remainder of your post you seem to say similar, but the way you put it you in the end say "nothing". You can do better.

Use something like this yourself and make some explicit sense out of it.

 

So in the context you put this remark (see first quote) I claim you make this up. Or hope it works.

Etc.

 

 

 

I hope that we gain access to the captures of the analogue out - I wouldn't be using that program, except out of curiosity ... so far, I have done exercises like this manually; I may find something, or I may not ...

[sandyk]

2 hours ago, manisandher said:

I think we all agree that -160dB, -130dB, etc., differences cannot be heard.

Mani

 esldude and others may believe that, but I certainly do NOT agree.

 

 

[manisandher]

Alex, there are two different case here:

1. using an FFT to 'see' under the noise floor

2. using a diff program to compare two files

 

We're talking about the latter case where Peter has already explained what's going on:

 

1 hour ago, PeterSt said:

We see noise only. We compare noise only. The guys of fame will tell you that you won't be able to hear differences at that level. Whether 160dB or 120dB. Noise is noise. But it isn't.

 

Mani.

[PeterSt]

44 minutes ago, manisandher said:

 

 

 

But about that one and for those who feel in the mood today (and for me for the 26th time) :

 

This is the bit perfect analogue result and registration of the difference between "a" software player which shows a running time cursor (this cursor jumps forward each second), vs the software player not showing/having that. You see the cursor in the bottom-middle here :

 

 

Both players are XXHighEnd but this is not important. It is the sheer fact that what all needs to happen to move this cursor draws current and this is what we can observe audibly. Remember, XXHighEnd has been explicitly developed to produce the very best SQ and it is about these matters, which can be measured just the same.

Don't forget : bit-equal for all outputs.

 

What literally happens here is that we have one disconnected Sound Engine (for insiders, XXEngine3.exe) which is a background task and picks up what to play from a controlling GUI (the XXHighEnd.exe program). One of the features is the quite important mode to play so-called "Unattendedly" which means nothing more or less than that the GUI (program) disappears fully (so it is nowhere in memory left - it just quit).

Its counterpart "Attended" mode, lets this GUI remain for real time control, and all what we see is actually simulated. So the running time is a virtualization from reality and the reality is within XXEngine3.exe which is the really sound producing program. Notice that XXHighEnd.exe itself could also produce the sound just the same, but now the GUI (program) would not be able to detach from the sound producing program.

 

All 'n all what is important to understand that it is just a lousy stupid moving cursor already which influences sound. The "how" you can just see in that plot. Easy enough. Well not really, because it takes (took me) two months to make the software and a virtually impossible setup to let it work for real. But I managed, back at the time.

 

If a knot in the interlinks would produce an audible difference (and as we know people really do such things) then this would show that difference. Would it NOT show any difference then or it is in the noise for real, or the difference does not exist (and we'd be placeboed by a knot)

 

While this shows XXHighEnd vs XXHighEnd in two modes, the various other dials in there also show the differences easily, but all in a different fashion. This is obviously also how I could learn what to do in the first place. Or, how Foobar will sound different than XXHighEnd. Etc. etc.

 

Put your teeth in this one to get the hang of it :

This is a time span of 0.7 seconds and a virtually continuous quite severe deviation from a reference which again is the red line (thus two recordings resulting in bit-equal output, but with continuous different analogue output). It shows a resonance of a buffer (size). So the size of the buffer influences itself.

That's how we do it. 

 

[fas42]

1 hour ago, PeterSt said:

 

And that is exactly the problem. Those differences ARE there as a fact and you can't do a thing with it for that reason. This is apart from the impossibility to align two different takes (only infinite ADC sampling rate would allow for that (so Mani, now you know).

 

Not true. It is possible to align the takes, but it's not trivial to do - I got halfway through the exercise of achieving an alignment sufficiently precise, but lost the motivation at that point; if the interest is there for this exercise, then it will be taken further.

[PeterSt]

1 minute ago, fas42 said:

Not true. It is possible to align the takes, but it's not trivial to do

 

Frank, then you didn't get what I was talking about.