Blue or red pill?

[esldude]

57 minutes ago, PeterSt said:

I know this is too quick and dirty but say it is a teaser. Try to open the below on a monitor of 3320+ pixels wide, or two next to each other, 1680+ each. Size is 3311x1006. Try to look at it at full size, as the dots shown take into account the monitor's resolution. For the general idea it is not thaaat important but when you want to pick nits, then it is. Literally.

 

This is the difference between the #15 and #16 files. #15 is the reference, #16 is shown as the difference against that.

 

Horizontally this spans just over 1ms, see the scale in the top (we're at 6.1 secs here). Vertically it is sample values (the middle line being 0), but divided by 256 because otherwise it goes off screen. So might you be able to count a span of 100 pixels vertically, this represents a value 25600 on the 16 million, assumed that Mani indeed recorded at -0dBFS and in 24 bits.

This is just one (1ms) part of the file of 28000ms (28 seconds) and you can stare at it for 20 minutes and still discover new things. Do notice that while a part you see here will be repeated elsewhere - say 2 seconds later -, e.g. 5 ms down the line this pattern changes (not shown but trust me).

 

 

You may like to concentrate on the first larger peak above zero, which emerges at "096" on the second (sub)scale (this is at second 6.099, the "100" being 6.100). You will see quite similar peaks each ~0.043 second, the second one appearing just before the 140 mark. 3rd just beyond 182. You can see from the rest of the plot that indeed this is a repeating pattern all over.

The band of "noise" is to be regarded the real environmental noise, not related to the subject. Anyway you can see that the width of the band of noise is quite similar all over (you could calculate the dB of it by means of seeing the width of the band in pixels and multiply by 256 and divide that on to 16million and convert to dB).

 

The orange graph is experimental and it heads towards identifying the patterns better. For example, our three peaks come along with 7, 8 and 7 downward excursions respectively and with that you may identify them as "almost equal". Might this pattern occur again 5 seconds further down the line, then the orange graph helps identifying it.

 

Because a pattern like the one we see is is not happening out of the blue, in Mani's system the 0.043 second interval should be recognizable one way or the other. Now, since the only parameter changed was the Split File Size (SFS) and we know that the one was set at 200 and the other at 0.1 (which numbers are without unit except for "SFS" itself) we can aim for the SFS of 0.1 interval to see what causes it. This, with the notice that the SFS of 200 will imply processing maybe once per minute. Now, if Mani would again play something of the same bit depth and sampling rate and with the same further settings (and SFS at 0.1 !), including the same upsampling rate and filtering, his logging may show something of the 0.043 second. I can not guarantee this because I can not do over here what Mani does behind his two closed doors, but he can put up the X3 and X3PB log files of this (not XX). I reckon that the 0.043 is a sub-happening in the midst of the split file part loading itself because 0.043 seconds will be too short for the implied chunk reading which may happen at an interval of say 20 times per second at this upsampled rate of 192000.

Don't worry about these numbers, but this is the way to approach a thing like this, if we want to know what happens in the first place.

 

Peter

Peter, 

 

Can you show the same kind of graph from the same exact time from the original digital file as a way to compare to the captures? It is the reference signal anyway.

[manisandher]

9 hours ago, Ralf11 said:

the affirmative has the burden of proof

 

Sure.

 

In this case, the affirmative is speculation about any potential 'tells' during the ABX. Mans says he has some evidence of a potential tell. I've said put up or shut up. His silence speaks volumes.

 

Mani.

[manisandher]

9 hours ago, pkane2001 said:

I'm starting to get most of the functionality I want in my own diffmaker-like software.

 

1 hour ago, PeterSt said:

I know this is too quick and dirty but say it is a teaser.

 

Wow. Really nice guys. Let me know if there's anything you'd like me to try.

 

1 hour ago, PeterSt said:

I can not guarantee this because I can not do over here what Mani does behind his two closed doors, but he can put up the X3 and X3PB log files of this (not XX).

 

I'll be able to do this over the weekend.

 

Mani.

[PeterSt]

1 hour ago, esldude said:

Can you show the same kind of graph from the same exact time from the original digital file as a way to compare to the captures?

 

Dennis, the software has been made for that too but it is the more difficult part because it now requires level matching as well (Mani's two recordings were of the exact same level to begin with). And, because I exactly left out that part for the 16 to 24 bit transition of the software (which again makes it a quick and dirty though now internally), this will be a manual adjustment - albeit doable.

 

What will not be doable as such is the different sampling rate and bit depth; the original file will be 16 bits and 44100 IIRC, which makes it more understandable why the software was 16 bits to begin with. This is what the originals mostly are (or were back in 2009 when this software was made), and the playback of it was not upsampled as were the recordings just in 16/44100. And with connected clocks ...

 

The real answer is : not doable. But :

I now realize that while you ask for the "original", there's also the digital capture; this is the upsampled version in 24 bits and I assume 176400 (because XXHighEnd does not really like uneven upsampling for the best results) with only the problem that the recordings are in 24/192000 if I am not mistaken. It should be no real issue for Mani to do it once again plus it also would not violate the ABX happening by any means, because that was unrelated to the recording and just played the 16/44100 at 24/176400 (I think).

So it would require the tracks to be played again and recorded by the Motu, the Motu recording at 176400. But why do I have in mind that the Motu can't do this ...

So I think that is a problem after all.

 

Stuck again ?

 

 

[Audiophile Neuroscience]

2 hours ago, Audiophile Neuroscience said:

 

Quote

Ralf11 said:

does the avatar help?

 

No. C'mmon Ralph, out with it !

 

Hmmm I have two working theories:

 

You are the mummy from space.

OR

An electron microscope image of the inner ear

[manisandher]

2 hours ago, PeterSt said:

So it would require the tracks to be played again and recorded by the Motu, the Motu recording at 176400. But why do I have in mind that the Motu can't do this ...

 

Hey Peter, just a very quick response for now. The MOTU can do 176400, but I had problems getting the software recorder to behave itself. I'm sure I can figure this out. Will do so when I get back home.

 

Mani.

[PeterSt]

22 minutes ago, manisandher said:

I'm sure I can figure this out.

 

Mani, great. But only if you have some time for it.

 

Peter

[adamdea]

6 hours ago, Audiophile Neuroscience said:

 

lets agree to disagree on how science works. What we have is scientific evidence to a p=0.01. That evidence was gained from a pre-agreed and reasonable test methodology. There is speculation as to confounding issues but no evidence to support the speculation. The current available evidence stands. It is not proof. It is scientific evidence. Conclusion : <quote>"there appears to be a hearable difference (by some people)"</quote>. Agreed.

 

18 hours ago, manisandher said:

 

Stop deflecting.

 

 

You say you've done the analysis. Where's your evidence for "potentially some skew" in the ABX? Put up or shut up.

 

Mani.

 

7 hours ago, Ralf11 said:

No, this not the way science works!  We have some evidence, but it is no way at scientific level.

 

I'll agree there appears to be a hearable difference (by some people).

 

 

 

Both sides of this argument seem to be ducking the obvious point. Any fule no that

- repeatability is what is required to turn a single experimental result into solid grounds to modify existing theory (not mere speculation about its significance either way).

- if anyone has a theory as to the existence of a confounding variable then that can and should be built into repeat experiments.

So the problem is -is there any real appetite it, this time checking to see whether the system does in fact produce any tell between a setting change and a non setting change and introducing a control to allow for that- more careful control over the pauses on each change to equalise or randomise them, possibly doing a settings change between each AA and A/B, whatever. Not saying there should be that appetite, but evidently that would be the way to move things on.

 

The argument about the evidential status of the single experiment in the absence of repeat is dull and fruitless. If anyone expects it to be taken as paradigm changing on it's own- go ahead and approach a scientific journal.  It is what it is, but what it is is not very much.

 

For my own part I think that an ADC  recording should be able to capture anything audible and then it can be ABX'd -providing instant repeatability (or not). But it seems that no one is very interested in that. The live software stream is much trickier to control and repeat. There we have it.

 

[mansr]

5 hours ago, manisandher said:

His silence speaks volumes.

Has it occurred to you that I might be busy?

[Audiophile Neuroscience]

12 minutes ago, mansr said:

Has it occurred to you that I might be busy?

 

Not too busy to make unverified speculations

[Audiophile Neuroscience]

58 minutes ago, adamdea said:

 

 

 

Both sides of this argument seem to be ducking the obvious point. Any fule no that

- repeatability is what is required to turn a single experimental result into solid grounds to modify existing theory (not mere speculation about its significance either way).

 

 

Well, not really. I have always said a reproducible result would strengthen confidence.Given the ungracious reception of the initial result I think the daffy duck "OH, NO not again" applies.

 

[mansr]

I looked a little closer at the timing of the playback. The gaps preceding the known A samples are, aside from one outlier, consistently shorter than those preceding the known B samples by an average of about 7 seconds. For the X samples, the preceding gap, however, shows no obvious correlation with the selection. Both extremes were B samples, and the mean differs by much less than a standard deviation. In light of this, I find it unlikely that the timing skewed Mani's choices towards the correct.

 

Happy now?

[Audiophile Neuroscience]

20 minutes ago, mansr said:

Happy now?

 

It is what it is. For me, I am happy to get to the truth, whichever way it falls.

[manisandher]

2 hours ago, mansr said:

In light of this, I find it unlikely that the timing skewed Mani's choices towards the correct.

 

Nothing skewed my choice of the correct other than a change in the sound of the music.

 

2 hours ago, mansr said:

Happy now?

 

Why would I be happy that you've shown something I already knew? It'd be good if you could stop positing unfounded speculations though.

 

Mani.

[mansr]

There's really no pleasing you, is there?

[mansr]

1 minute ago, manisandher said:

Nothing skewed my choice of the correct other than a change in the sound of the music.

I believe that you believe that.

[look&listen]

7 hours ago, Audiophile Neuroscience said:

 

Not too busy to make unverified speculations

Predicted long ago in thread.

Maybe helpful - https://theconversation.com/how-to-reason-with-flat-earthers-it-may-not-help-though-95160

[Ralf11]

12 hours ago, Audiophile Neuroscience said:

 

Hmmm I have two working theories:

 

You are the mummy from space.

OR

An electron microscope image of the inner ear

 

# 2 - discussed on another thread - sending you a pm

[Ralf11]

It would be easy to test the clicking thru closed doors tell hypo. as a confounding effect.  Mani could kill that one w/out much time investment.

 

I do not think this has a very important effect on SQ, and would much rather see you guys engineer some great (and cheap) DSP for my Maggies.  But (BUT) it is interesting, and I will contribute a beer* if you or even any one of you plus another person does another trial - maybe dust off that phasure next time(?)

 

 

* transportation costs not included

[mansr]

2 minutes ago, Ralf11 said:

It would be easy to test the clicking thru closed doors tell hypo. as a confounding effect.

For the record, I seriously doubt that was happening. I mentioned it only as an example of things that could, in some setting, have influenced the outcome. I remember reading about an ABX session that was compromised because the test subjects were able to hear a difference in the clicking of a relay, so such things do happen.

[Audiophile Neuroscience]

4 hours ago, mansr said:

For the record, I seriously doubt that was happening. I mentioned it only as an example of things that could, in some setting, have influenced the outcome. I remember reading about an ABX session that was compromised because the test subjects were able to hear a difference in the clicking of a relay, so such things do happen.

 

I am not doubting this but I would have thought the subjects would need to be aware of the clicking to make it a useful "tell" and make a conscious differentiation between A and B. All happening at a subliminal unconscious level seems a bit out there to me. This, to me, is different to gravitating to a choice due to a preference which is hard to put your finger on or otherwise 'unaware' of.

 

Anyway @manisandher should maybe consider the suggestion of @Ralf11 to eliminate it as a confounder. I really think it is far fetched though.

[Ralf11]

We do know that things at a subliminal unconscious level can affect choices, preferences and perceptions - even in cross-sensory areas.

 

in ~2010 German researchers tested preferences and perceptions of 150 subjects drinking Riesling wine - they preferred the wines under red or blue light rel. to green or white light

 

135 subjects (hapless undergrads. IIRC)  perceived 8 wine samples as 'spicier' under green or blue light and fruitier under red light

 

Cambridge Univ. has an entire lab center on cross-sensory effects

 

I'll agree to seemingly unlikely on the keyboard tells thing, but if you have an unusual result you want unusually good evidence.  I suggest you guys continue, expand and publish.

[Sonicularity]

From the 10 ABX tests alone, we can deduce that switching from B to A is identifiable.  Even on the test where the choice was changed, it was changed only after switching between B and back to A.  It was B, but when the test was started over, A was played again, and then the "1?" that was written down was changed to a "2".  

 

Why not eliminate the potential issue when switching from B to A?  Try repeating the test with a BAX order.   This way, X is always played immediately after sample A rather than sample B.  

 

 

 

 

 

 

 

 

 

[manisandher]

4 hours ago, Sonicularity said:

From the 10 ABX tests alone, we can deduce that switching from B to A is identifiable.

 

Uh, no.

 

The single response I got wrong was my response to the 9th ABX where I thought it was '1', but it was actually '2'. I.e., I thought there was a switch from B to A, but there wasn't. So switching from B to A is not identifiable... other than in the sound of the music. (Why I got the 9th ABX wrong remains a mystery to me - we were ~15 mins into the ABX, so perhaps it was just down to being tired.)

 

And in any event, I have shown that there were no 'tells' in the two failed ABXXXXXXXXXX tests. If you believe there might have been tells in the subsequent 10 ABXs, then they must have magically appeared.

 

I don't believe in magic.

 

Mani.

[Audiophile Neuroscience]

1 hour ago, manisandher said:

 

 there were no 'tells' in the two failed ABXXXXXXXXXX tests. If you believe there might have been tells in the subsequent 10 ABXs, then they must have magically appeared.

 

Yes, whether conscious of them or not, the same tells if they existed would have been in operation during the previous ABXXXXXXX tests. If present they would have told you the correct responses. Since you failed the ABXXXXXXX tests there could not have been any tells and therefore none carried forward into the successful ABX trials.