Someone just posted ABX results on reddit, that shows they were able to successfully ABX an 88.2 files against the same file converted to 44.1.
I have yet to hear of any "hi-res" music passing an ABX test against it's CD Quality countrpart.
Here's a link to the post:
https://www.reddit.com/r/headphones/comment...s_out_a/cuv8866 (https://www.reddit.com/r/headphones/comments/3k3cmg/business_insider_gets_a_fiio_e6_freaks_out_a/cuv8866)
I don't understand how this is even possible.
He said he downsampled with Gold Wave ... have a look at the two entries for it profiled on http://src.infinitewave.ca/ (http://src.infinitewave.ca/). One version is OK, but the other, older one... tons of aliasing, probably audible.
Also unfortunately foo_abx can't currently be used to reliably ABX files that have different sample rates. The delay when switching from sample A to X is longer if X doesn't have the same sample rate so track identities are revealed by accident.
Until this problem is fixed the user should upsample the new file back to the original samplerate when ABXing the difference.
Another thing worth to point out is the test is done using ASIO. Without any SRC in the DSP chain the playback device need to switch its clock when playback at different sample rates. The switching process can make a noticeable delay and even a click/pop noise.
Moreover, the playback device can actually perform differently in different sample rates therefore the result is device-dependent rather than the format itself.
Also, intersample peaks in the source file can result in clipping after resampling, so it is a good idea to reduce about 2-3dB while testing unless the source file is already quiet enough to avoid intersample peaks.
Assume there is no deliberate cheating, the more appropriate way to perform the test is use replaygain to reduce the volume of both files and use a DSP resampler like the SoX plugin to upsample the 44.1k file to 88.2k again to avoid switching noise or delay while testing, and avoid device-dependent sound quality differences as well.
I'm not a professional or anything, I just know what I like and generally prefer HD audio vs the standard. It's possible that the mastering is different in many HD releases and that's why I prefer them, but I'm not certain. I will say that I've ABX'ed 24 and 16 bit audio and couldn't tell a difference, getting a whopping 9/16 attempts correct, but i do hear something different between sufficiently different samplerates. It's quite possible that downsampling in goldwave produced audible artifacts on its own, but this was just something I did for the fun of it about a month or so ago. I'm not trying to say I've got magic ears or anything, but I do often hear a difference, even if I can't pinpoint it other than to simply say that it sounds 'better'.
Edit:
The track was resampled using Goldwave 6.13
> I do often hear a difference, even if I can't pinpoint it other than to simply say that it sounds 'better'.
And that's why we've got ABX.
Am I correct in understanding the actual files he used haven't been posted anywhere for others to download, evaluate, and attempt to ABX themselves?
Provided solely for the purposes of research and critique: https://mega.nz/#!PMp1nS6A!jvyTUhhE...zfCGeJE4eg3Dojg (https://mega.nz/#!PMp1nS6A!jvyTUhhETu1eIfOn772oNrZ6xzsXzfCGeJE4eg3Dojg)
Provided solely for the purposes of research and critique: https://mega.nz/#!PMp1nS6A!jvyTUhhE...zfCGeJE4eg3Dojg (https://mega.nz/#!PMp1nS6A!jvyTUhhETu1eIfOn772oNrZ6xzsXzfCGeJE4eg3Dojg)
Just to let you know, I didn't post this here to call you out. It's just that these guys know WAY MORE about this stuff than I do.
I spent weeks ABXing 24/96 vs downsampled 16/44.1 and couldn't tell a difference on about 2 dozen tracks. I had to stop because my wife was getting sick of seeing me on headphones all the time.
So, I was curious what could be the cause of your golden ears. I'm still planning to ABX the files you send me on Reddit yesterday when I have a 15 minute break during the day.
A general ABX question. All ABX shows you is that you can tell the difference between 2 files. It doesn't actually tell you which one you like better.
So it may be possible that you subjectively likes a high res or a FLAC file better than it's low-res/lossless counterpart.
But if you successfully can distinguish two files via ABX, it would be nice if the tool let you blindly listen to clips from the track and pick which one you think you like better.
I'm also gathering from this thread, that it's really not possible to properly ABX a high res file against a standard 44.1/16 FLAC due to limitations of foobar2000 and foo_abx plugin? Is that correct?
But if you successfully can distinguish two files via ABX, it would be nice if the tool let you blindly listen to clips from the track and pick which one you think you like better.
ABX has the specific goal of establishing if there is/is not an audible difference. If what you want is to determine a preference then that is a completely different tool, ABC/HR.
Obviously if you fail to hear a difference with ABX then there is no need to use ABC/HR.
But if you successfully can distinguish two files via ABX, it would be nice if the tool let you blindly listen to clips from the track and pick which one you think you like better.
ABX has the specific goal of establishing if there is/is not an audible difference. If what you want is to determine a preference then that is a completely different tool, ABC/HR.
Obviously if you fail to hear a difference with ABX then there is no need to use ABC/HR.
Is there an ABC plugin for foobar2000? I would love to ABC different remasters of some older albums.
This is in reply to your previous post.
In post #3, Case already addressed the other question. Yes it is possible and it's a bit of a stretch to call upsampling the lower res track improper, unless you use a junky SRC. I'd probably not use Gold Wave just to be on the safe side.
It seems that jdeberhart's 44.1k file is properly resampled. There are only very few possibly clipped samples in the 44.1k version, other than that it generally nulled with the 88.2k version down to -140dB up to 20kHz when I upsampled it to 88.2k again using the SoX plugin.
Then the main problem left is the delay when switching sample rates. This video clearly demonstrates how to exploit the delay to know the correct answer, and how a DSP resampler can eliminate this delay.
https://youtu.be/H6aRV3ZZWiE (https://youtu.be/H6aRV3ZZWiE)
Download the SoX resampler plugin here:
https://www.hydrogenaud.io/forums/index.php...ost&id=7331 (https://www.hydrogenaud.io/forums/index.php?act=attach&type=post&id=7331)
I don't know that it rules out the possibility of samplerate-specific hardware coloration.
Just lately we had a member abxed positive his behringer DAC with 88.2->44.1 but not back to 88.2.
I don't know how the Valhalla tubes in the playback chain poduce IM with and without hf content.
Also i don't take any abx of high against low samplerates with foobar to serious anymore because cheating is so easy with even a bad spectral analyser running at the desktop. Tis may of course only show how much of a badbass me is.
I'm a little concerned that we're coming up with excuses why this test passed, when we feel it should not have passed.
If there are specific steps that the community here feels are needed to be followed, in order to properly ABX hi-resolution music vs CD Quality, I'm thinking we need to document them somewhere (perhaps the Wiki) and justify the sh** out of why we require this, or the Stereophile Magazine guys will have a field day with us.
My apologies for violating TOS #6 with the title. Thank you to whatever mod fixed my oversight.
I'm a little concerned that we're coming up with excuses why this test passed, when we feel it should not have passed.
meh
Stereophile Magazine guys will have a field day with us.
meh
Explaining the pitfalls of homebrew testing is fine, but trying to explain what constitutes good science to those who pretend to embrace it only when it suits their agenda?
I'm a little concerned that we're coming up with excuses why this test passed, when we feel it should not have passed.
meh
Stereophile Magazine guys will have a field day with us.
meh
Explaining the pitfalls of homebrew testing is fine, but trying to explain what constitutes good science to those who pretend to embrace it only when it suits their agenda?
I don't want to give those guys any talking points against people that get it. I guess I'm just cynical that way.
You're not cynical enough, I'm afraid.
I'm a little concerned that we're coming up with excuses why this test passed, when we feel it should not have passed.
If there are specific steps that the community here feels are needed to be followed, in order to properly ABX hi-resolution music vs CD Quality, I'm thinking we need to document them somewhere (perhaps the Wiki) and justify the sh** out of why we require this, or the Stereophile Magazine guys will have a field day with us.
We have seen far too many false positives reported over the years to take anything for granted.
It's not as though the regulars here haven't tried to ABX high sample rate vs. 44.1, just that none have reported success, and a lot of these folks are extremely good at detecting small differences. They are also very good at understanding the pitfalls inherent in properly performing such a test.
It seems that jdeberhart's 44.1k file is properly resampled. There are only very few possibly clipped samples in the 44.1k version, other than that it generally nulled with the 88.2k version down to -140dB up to 20kHz when I upsampled it to 88.2k again using the SoX plugin.
-140dB is well below the capabities of 16-bit which jdeberhart says is transparent to him, so at this point it should be clear to him that something else is going on. I'm giving him the benefit of the doubt that he isn't cheating.
this (https://www.hydrogenaud.io/forums/index.php?s=&showtopic=110058&view=findpost&p=906748)
The log says foo_dsd_asio as output. Does this mean your PCM files are converted to DSD before being sent to your DAC? That would be another variable that could cause audible differences between different sampling rates.
And could you please run one of these high frequency impulses at normal listening volume and tell us if you hear anything, jdeberhart?
test files (https://www.hydrogenaud.io/forums/index.php?s=&showtopic=107588&view=findpost&p=882142)
I tried the impulse test and couldn't hear anything. I'll try ABXing the 44.1 upsampled back to 88.2 this weekend when I've had some rest, though an initial listen isn't promising, leading me to think that there's some sort of samplerate-specific filtering going on in the DAC, which given that it's an ES9018K2M is quite possible.
I'm using the DSD output only for DSD files though, PCM is not converted and instead played back 'native', though the 9018 is still a sigma-delta DAC so ultimately everything is getting converted anyway.
Maybe the differences are from the tube amp's IM as Wombat stated so you may want to remove or change it to something else as well.
https://www.reddit.com/r/headphones/comment...s_out_a/cuv8xf5 (https://www.reddit.com/r/headphones/comments/3k3cmg/business_insider_gets_a_fiio_e6_freaks_out_a/cuv8xf5)
Some additional data is outside of human hearing range, but high frequency distortion is reduced at higher samplerates. Listening for this distortion is where i was able to pick out the differences.
In case of IM it is actually the opposite. The "192kHz considered harmful" section in this article explained the effect with sample audio files for you to try.
http://xiph.org/~xiphmont/demo/neil-young.html (http://xiph.org/~xiphmont/demo/neil-young.html)
I'm a little concerned that we're coming up with excuses why this test passed, when we feel it should not have passed.
Seemingly a valid point - consistently invoking ad hoc hypotheses, that is denialism 101. But the reservations made here are issues well-known beforehand, that's a big difference. (Of course the typical antiscience-denialism site [insert here anti-vaccination/young earth-creationism/climate change denial/tobacco isn't harmful/six million lies] will have its own "well-known" reservation readymade too, but I suppose that a scientist from outer space would be able to ABX out hydrogenaudio from those at about 15 out of 16 anyway. Well they would likely look for conspiracy theories rather than adhoc-ness ... oh, whatever.)
Part of the problem is that we have no falsification test, and then something that is only a valid (up to chance outside the confidence interval) verification test if done right. That of course is an improvement over a verification test which is known to lead to false positives because the (sighted) environment is
inherently flawed - those who fail to see the problem with that setup, will never need to resort to ad hoc hypotheses over the flaw anyway.
Now. If one can believe the arguments that ultrasound could induce harmful distortion (edit: bennetng beat me to it),
why should anyone ever be surprised to see 88.2 ABXed from 44.1 even if the latter is upsampled to avoid switching issues?
I'm a little concerned that we're coming up with excuses why this test passed, when we feel it should not have passed.
That is the usual audiophile suspicion. One gets used to it.
The actual problem is that it isn't very useful to know that there was an audible difference. You also need to know
why there was an audible difference, otherwise you can't reach a conclusion, or find an explanation. And often, you can't confidently say that your test setup was good, when you have no idea why you got the result you got. This sort of problem is common in science. You conduct an experiment, you have some idea what the result ought to be, and the experiment is expected to confirm it. If it doesn't, and the results are surprising, that is not treated as a disproof of your hypothesis. It is something that needs a proper explanation, and the explanation could well be a flaw in the experiment. Only when you can't find one, and an independent experiment yields a similar result, you're on to something new. It is still not disproving anything, it is something that appears to contradict established knowledge and lacks a proper explanation. Before letting go of established knowledge, the next step would be to try finding an appropriate explanation, which removes such contradictions, and doesn't create even more of them.
For an example, look at the highly publicised experiment at CERN a couple of years ago, where it seemed for a while that they had broken the speed of light barrier, only to find out after a while that they had a flaw in an optical connection somewhere in their setup. There were people who said right from the start that they suspect a flaw somewhere, and that they wouldn't bet on the result to hold up. You could have told them the same: That they are seeking excuses for an unwanted result.
May I add my $0.02 to that.
I'm not a complete newbie when it comes to ABX testing, and yet I made mistakes in the past that lead to positive results.
Now if given the benefit of the doubt (no cheating, no lucky guesses) such a test result simply means that there most likely was an audible difference. But sometimes after recreating the test files properly and minimizing uncontrolled variables this difference disappeared.
So I think it is only fair to try to find out more details about this particular test, try to minimize uncontrolled variables until we're ideally just left with what we really want to test and try to reproduce the results.
This has nothing to do with excuses.
I tried the impulse test and couldn't hear anything. I'll try ABXing the 44.1 upsampled back to 88.2 this weekend when I've had some rest, though an initial listen isn't promising, leading me to think that there's some sort of samplerate-specific filtering going on in the DAC, which given that it's an ES9018K2M is quite possible.
I'm using the DSD output only for DSD files though, PCM is not converted and instead played back 'native', though the 9018 is still a sigma-delta DAC so ultimately everything is getting converted anyway.
Thanks for the short report. If you won't be able to tell the 88.2 reresampled music we may have the second person in a short time proving different DAC sound at 44.1 against higher samplerates.
I really wonder if this comes from attempts of the DAC manufactors to implement funky filters the audience asks for or purposely preventing to risk a bad reputation if it doesn't clearly sounds better with higher rates.
If some DACs do weird extra things for different sample rates, shouldn't it be easier to just measure it, instead of doing fragile listening tests?
and try to reproduce the results.
Of course, there will always be
some false positives reported by chance, even without any technical flaws. Trying to reproduce would then give another false positive in one case in twenty as well (or whatever your p-value threshold dictates).
Yes, of course, statistically another case can be made to be careful but I didn't want to go there.
Regarding the sample rate dependent sound quality issue, a famous example is Creative 10kx soundcards (Live/Audigy) having very different measured results in 44k vs 48k.
Another not so famous example is Asus Xonar D2 (the first Xonar). The measurements in ixbt and even the official Asus pdf report showed obviously more inferior performance in 44k, even though it is unlikely to be audible.
http://ixbtlabs.com/articles3/multimedia/asus-d2.html (http://ixbtlabs.com/articles3/multimedia/asus-d2.html)
See page 14-15
http://audio.rightmark.org/downloads/Xonar...stGuide_V12.pdf (http://audio.rightmark.org/downloads/Xonar_D2_RMAA605_TestGuide_V12.pdf)
The result of Xonar D2 in 44k is strange because unlike Creative's cards, Xonar D2 is supposed to support 44.1/88.2/176.4k clocks natively, as shown in the specs
https://www.asus.com/Sound-Cards/Xonar_D2PM/specifications/ (https://www.asus.com/Sound-Cards/Xonar_D2PM/specifications/)
I mostly think about options like Slow Roll-Off filters that a modern DAC chip has to have. It fights the terrific ringing! These filters can damp highs a lot. People with good hf hearing may clearly hear that while they never will know what that nasty ringing is all about.
The roll-off doesn't matter with high samplerates.
The DAC chip used in this test is capable of this Slow filter.
Nasty, terrific ringing. Some here might not know that you're being sarcastic.
Nasty, terrific ringing. Some here might not know that you're being sarcastic.
Sorry, i can't help myself when touching that topic.
Creating the fear of ringing was surely one of the best marketing strategies in audio business ever. It created the need for so many solutions against it that can only be described as terrific. Only dsd can rescue us!
Something like this?
http://archimago.blogspot.hk/2013/06/measu...ilters-and.html (http://archimago.blogspot.hk/2013/06/measurements-digital-filters-and.html)
If it is audible why the reason is ringing, but not high frequency rolloff or IMD or aliasing?
Thanks for the link. This is again a nice one from Archimago. I should gave remembered that myself.