Why 24/192 is a bad idea?

[jkeny]

Well... we don't really have any one way or the other. Eg, how did you know the choices were correct without looking at the playlist?

Simple after I marked my choices I went to the computer & looked at the list of what just played. Started another random run & repeated. Not difficult

Six or seven doesn't say much (I just now sat here hammering the 'A' key in 15 successive runs and got 7/7 twice and 0/7 once. Statistics.... dude), and then there's trusting others to report faithfully. There are plenty of ways this goes wrong, the same way that you don't get a random sequence by having people pick marbles out of a bag.

I know but I never presented it as statistically rigorous - I'm not pretending to do fully fledged DBX testing.

Heh. The first thing I did was look at the files in an analyzer. I would suspect more than a few reporters couldn't resist either. Even Tiefenbrun couldn't resist using any cues available to be right, he just got embarrassed/caught because he remembered the relay noises backwards!

I know at least half of the people who did this test & know that they did not analyse the files first. I specifically asked you or Arny to listen first & look later. A bit like Ethan's reply - he subsequently wanted knowledge about the files & I asked why? SO you guys really don't trust your ears - even though you proclaim blind testing for others, you won't abide by this yourself?

The file sizes make guessing easy in retrospect too.

Yes if you want to peek but as I said I & others trust our ears & don't tend to peek . This is hilarious, I ask you to do a blind test & you try to cheat

In short, you don't have any reason to trust me completely, and I don't have any reason to trust you completely, and neither of us should trust random reporters on the net especially if they're not even pretending to do a fully blind test. Collecting real data is doomed in an informal setting, so we have to settle for discussion and being content knowing how rigorous we ourselves are being.

Come on man, you'r reaching! I know it's human nature to try defend your position especially when your ears or audio equipment is called into question but the results speak for themselves. Not rigorous, no but this isn't AES either The reports from the people who heard a difference was so like each other that no explanation you have given so far would explain this unless you want to say that they collaborated?

I would think the difference between jplay and foobar would be a interesting investigation?

It's already been done & Jplay wins hands down!!

 

[rbbert]

I have received (anonymous & indirect) feedback that my posts in this thread appear to be unwelcoming to our newest industry members, Monty and jmvalin. Having looked back and reflected on it, I think the criticism is well placed and I apologize for that to both of them and members who were so disappointed in my interactions with them. As much as I try to have control over my emotions in these topics, some amount of it gets pulled in, in the heat of the argument. On WBF there is no excuse for that. These topics can be very complex and opinions and interpretation of science certainly polarizing to an extreme. All positions are welcome on WBF in these matters and nothing in me answering them should be taken as otherwise.

I wish Monty and Jmvalin would continue to contribute to this thread and elsewhere and find WBF a productive and comfortable "home" to have such discussions. Should they choose not to, it will be a source of severe disappointment for me personally.

Finally, you may want to know about an informal rule of mine: to the extent I am engaged in any discussion/argument with members, you will never see me exercise my forum powers to sanction my counterparts. Should an action be necessary, other members of our "management" team will be consulted and will make the final call, not me. So please do not fear any retribution and take comfort in me being no more privileged than you in this regard. Should you ever feel concerned, please feel free to contact Steve and voice it with him.

I think the idea of someone complaining about your posts is somewhat ludicrous. I haven't (in this topic at least) read that you are telling the rest of us what we can and can't hear and then repetitively dismissing (with little or no convincing evidence) allegations that "we" can indeed hear these things (also sometimes with little or no convincing evidence, true, but also sometimes with some evidence)

 

[jkeny]

Oh, and to be clear-- some level of preecho on this scale will eventually be audible, I'm not sure what the threshold will be. This is a true echo, though, not the kind of sinc spreading I think you were originally trying to illustrate? I've honestly lost track.

All I can say for certain: I personally don't hear it.

Yes, I guess this demonstrates that have no chance of hearing pre-echo or time-smearing from equiripple, linear phase digital interpolation filters?
There's a DAC chip, the PCM5102 that has both a equiripple, linear-phase filter & a minimum-phase filter on it which is easily switched between. The difference in sound between the two filters is very obvious & much more pronounced than the differences between these files. Using higher sample rates allows a relaxation in the filters which seems to result in the same type of improvement as is heard here - a sharper, less time-smeared & therefore more precise presentation. Is this the only benefit to using a higher sample rate, I doubt it but it seems to me to be a first order effect.

Edit: Oh & BTW, these files also show Winer's rules of transparency are wrong!
Off to bed now so any replies necessary will be tomorrow

 

[xiphmont]

Simple after I marked my choices I went to the computer & looked at the list of what just played. Started another random run & repeated. Not difficult I know but I never presented it as statistically rigorous - I'm not pretending to do fully fledged DBX testing.

Sure. Wasn't trying to suggest duplicity, just pointing out it wasn't rigorous.

I know at least half of the people who did this test & know that they did not analyse the files first.

Nod, and that does count for quite a bit.

SO you guys really don't trust your ears - even though you proclaim blind testing for others, you won't abide by this yourself?

No, we're curious. And because we are committed to using an ABX tool for real, we can play with the samples and then use a test that doesn't let us cheat. After all... you *made* the files, and you trust your own results, right?

Oh, and for the record, the test as you've constructed it-- I would not be at all surprised to find out that some people really can hear the difference. It's at the edge of both FR discrimination and critical band depth. But at this moment, I personally can't hear the difference.

Yes if you want to peek but as I said I & others trust our ears & don't tend to peek . This is hilarious, I ask you to do a blind test & you try to cheat

Did not! I used the ABX tool.

For the record, every human cheats. That's what big powerful brains are for. That's why the tools that don't let you cheat are useful.

But the results speak for themselves.

That's the only part that isn't quite so. They don't speak for themselves because you haven't reached the bar required for completely trustworthy evidence. I also didn't say you were wrong (I really do think in fact there's a good chance you're right), just that you haven't proven the case one way or another. And, to be fair, you weren't setting out to do that.

You've provided some additional weight to an interesting discussion, it's not right to say it settles it.

The reports from the people who heard a difference was so like each other that no explanation you have given so far would explain this unless you want to say that they collaborated?

Lots of people have 'independently' come up with the same description of little grey men too.

It's already been done & Jplay wins hands down!!

No curiosity at all? They were playing the same bits through the same hardware, how did the results come out different? You want us to trust your results when you have no interest in sorting out this fairly major discrepancy?

[or was my humor meter just miscalibrated?]

 

[xiphmont]

Yes, I guess this demonstrates that have no chance of hearing pre-echo or time-smearing from equiripple, linear phase digital interpolation filters?

[...]

Using higher sample rates allows a relaxation in the filters which seems to result in the same type of improvement as is heard here - a sharper, less time-smeared & therefore more precise presentation. Is this the only benefit to using a higher sample rate, I doubt it but it seems to me to be a first order effect.

Woah, woah, woah. Where do you get any of that? The samples you're testing have set up a preecho that has almost nothing in common, qualitatively or quantitatively, with sinc spreading (what you're referring to as "pre-echo or time-smearing from equiripple, linear phase digital interpolation filters"; it isn't time smearing at all. It's an inherent feature of any band limited signal. Filters do not cause it; it simply is.)

[edit: as opus111 has rightly j'accused below, I think I completely misread you here, so feel free to disregard the above. I'm leaving it up as public pennance.]

Edit: Oh & BTW, these files also show Winer's rules of transparency are wrong!

How so? You have .2dB passband ripple. His rules of transparency call for less than .1dB.

Monty

 

[opus111]

Woah, woah, woah. Where do you get any of that? The samples you're testing have set up a preecho that has almost nothing in common, qualitatively or quantitatively, with sinc spreading (what you're referring to as "pre-echo or time-smearing from equiripple, linear phase digital interpolation filters"; it isn't time smearing at all. It's an inherent feature of any band limited signal. Filters do not cause it; it simply is.)

Phew where did you get all of that? Pre-echo is by no means a feature of band-limiting, its a feature (as is explained by the Julian Dunn paper) of using certain methods (equiripple) to design linear phase digital filters. I can only suggest go back to that paper and digest some more Perhaps you're conflating pre-echo with pre-ringing (its a very common mistake to make).

In the paper the time smearing resulting from such digital filters is both pre- and post-echo, in jkeny's files he's exploring the audibility of just pre-echo.

 

[xiphmont]

Phew where did you get all of that? Pre-echo is by no means a feature of band-limiting, its a feature (as is explained by the Julian Dunn paper) of using certain methods (equiripple) to design linear phase digital filters.

Right, gibbs != preecho

I read what he wrote as attempting to extrapolate his test to all linear phase filters [and thus higher sampling rates regardless of filter design]. You may be right that I read that completely wrong.

Monty

 

[xiphmont]

I can only suggest go back to that paper and digest some more

Actually, you're right... I do want to go back and look at the paper in detail, as well as catch up with the MATHCAD files Don has forwarded me. I probably won't have time to do any empirical simulations or measurements for a few days though :-(

 

[NorthStar]

Hey Monty, what does it take to be friend with you? ...No emoticons? :b

Bob

EDIT: Ah finally!

 

[jkeny]

Thank you Opus & yes exactly correct!
Good Monty, I didn't know where you were coming from on this? I thought I was clear about what I was doing & why?
No, I wasn't talking about all filters, just the majority of filters used in digital audio at the moment


It is stated here that +/- 0.1dB as one of the definitions for WIner's audio transparency

Ethan Winer, an acoustics expert, discusses how audio electronics can be defined as audibly transparent by four broad categories of measurements and he provides his personal criteria for complete transparency. He states that gear passing all these criteria will not contribute any audible sound of its own and in fact sound the same as any other gear passing the same criteria:
Frequency Response: 20 hz to 20 Khz +/- 0.1 dB
Distortion: At least 100 dB (0.001%) below the music while others consider 80 dB (0.01%) to be sufficient and Ethan’s own tests confirm that (see below).
Noise: At least 100 dB below the music
Time Based Errors – In the digital world this is jitter and the 100 dB rule applies for jitter components.

Where does Winer state that it is 0.1dB (& not +/-)?

 

[jkeny]

I really have to laugh at some of you guys - you are so caught up in the blind testing bit that you fail to see the flaw in your own testing procedures.

Ethan Winer complained that I gave no information about the test files (I did btw, he just missed it) & I asked why? Would this have effected his listening?
You, Monty, tested the files before listening & then proceeded to declare that you used ABX & so couldn't cheat!

In both cases, you are setting up your expectation bias & this I may ad is exactly what you complain about in sighted tests. If your measurements show that differences are not enough to be heard then your expectation is that you won't hear any differences & lo & behold, you don't. Amazing!

I've seen this same behaviour from certain individuals on other forums also! I guess it's the fear of hearing something that measurements might not explain!!

 

[jkeny]

No curiosity at all? They were playing the same bits through the same hardware, how did the results come out different? You want us to trust your results when you have no interest in sorting out this fairly major discrepancy?

[or was my humor meter just miscalibrated?]

Oh god, here we go again!!
I'm not even going to open up this can of worms as it's been done to death but please retract your implied distrust & superior attitude shown. This is so commonly encountered by those who thinks bits is bits & that's all that counts. You have already admitted to your misunderstandings about filter pre-echo perhaps you could be wrong about this too?

 

[Phelonious Ponk]

Not following what amuses you, John, but when I went through component and media testing using blind listening the reason was very simple -- The difference between theoretically neutral and the measurements (if I have them) of the component I'm testing don't mean a thing if I can't hear it. And I don't pretend that my casual listening tests are scientific or completely devoid of all possible bias. But they do eliminate knowledge of the component I'm listening to, and I find that helpful. I may be a raging objectivist, but I'm not immune. When I have borrowed or ordered in a shiny, beautiful new DAC and I can afford it, I'm looking for an excuse to own it, just like everyone else.

Tim

 

[jkeny]

Not following what amuses you, John, but when I went through component and media testing using blind listening the reason was very simple -- The difference between theoretically neutral and the measurements (if I have them) of the component I'm testing don't mean a thing if I can't hear it. And I don't pretend that my casual listening tests are scientific or completely devoid of all possible bias. But they do eliminate knowledge of the component I'm listening to, and I find that helpful. I may be a raging objectivist, but I'm not immune. When I have borrowed or ordered in a shiny, beautiful new DAC and I can afford it, I'm looking for an excuse to own it, just like everyone else.

Tim

It's amusing that Monty can state that he can't cheat by using ABX when in fact he's already cheated by looking at the measurements. Not truly blind now is it?
Expectation bias is strong, either from sighted bias or from measurements bias. Eliminating sightedness & then substituting another expectation bias is what has me laughing!!
Winer is the same!
I don't see what you don't find funny?

 

[Clive]

John mentioned this thread to me as I asked why he wanted me to audition two files he sent. He's not asked me to post here but just want to say:

I had no clue what was going on before I heard the tracks and that the differences were really obvious immediately, I replayed them a few times but really I only needed a first play to hear the differences - John either cheated royally or the differences are easily audible with a good system!

My system has full range speakers and good resolution.

 

[jkeny]

Thanks Clive, for joining up & posting. Welcome!
And just to clarify - I told you this AFTER you had reported your listening results. I told you next to nothing about the files before the test, right?
Also, you listened to the 20 sample file & not to the one offset by 60 samples. The 20 sample offset is more difficult to hear.

John mentioned this thread to me as I asked why he wanted me to audition two files he sent. He's not asked me to post here but just want to say:

I had no clue what was going on before I heard the tracks and that the differences were really obvious immediately, I replayed them a few times but really I only needed a first play to hear the differences - John either cheated royally or the differences are easily audible with a good system!

My system has full range speakers and good resolution.

 

[Clive]

And just to clarify - I told you this AFTER you had reported your listening results. I told you next to nothing about the files, right?

Correct!

Also, you listened to the 20 sample file & not to the one offset by 60 samples. The 20 sample offset is more difficult to hear.

I dread to think how obvious the 60 version would be.....

 

[jkeny]

I dread to think how obvious the 60 version would be.....

Indeed, I can just about hear it on my laptop's internal audio through in-ear headphones

 

[Phelonious Ponk]

It's amusing that Monty can state that he can't cheat by using ABX when in fact he's already cheated by looking at the measurements. Not truly blind now is it?
Expectation bias is strong, either from sighted bias or from measurements bias. Eliminating sightedness & then substituting another expectation bias is what has me laughing!!
Winer is the same!
I don't see what you don't find funny?

I understand your amusement now. Yes looking at the numbers could create an expectation that there will be no audible difference and then you could hear what you expect to hear, nothing. No question about it. Unquestionably the best is DBT AB/X, when you don't even know what you're listening for, other than to identify whether X is A or B. Hard to expect anything then.

Tim

 

[Orb]

I understand your amusement now. Yes looking at the numbers could create an expectation that there will be no audible difference and then you could hear what you expect to hear, nothing. No question about it. Unquestionably the best is DBT AB/X, when you don't even know what you're listening for, other than to identify whether X is A or B. Hard to expect anything then.

Tim

Tim just using yours as it is the last post on this subject in the thread.

Can I suggest we leave the JND, double-blind testing,etc to the many other threads we have on such discussions.
You may remember I provided a lot of scientific papers highlighting the complexity of any testing including ABX that still suffers from AB order bias, and other biases similar to more traditional JDN double blind testing methods used in science outside the audio community.
Not going to say anymore on that subject here as I feel it is not the right place and it has been discussed many times already, but I have provided a lot of information in the past forums on this, including how with perception differences becoming more difficult then the number of passes reduce....
None of this is really relevant to this thread anyway as the focus should be processes-functionality-algorithms-etc with the consideration from the studio all the way to the consumer DAC (or anywhere in between including aspect of chip-DAC embedded/external architecture)
Thanks
Orb