Hi Folks,
For anyone who is interested in doing a direct real-world comparison between 24-bit, 16-bit and MP3, we are starting a little 24-bit download series via Twitter.
First up is some awesome Jazz from Alyn Cosker.
Download the full 16bit and 24bit FLAC versions of this track for free and have a listen. I think you will be pleasantly surprised! Click the little down arrow on the right hand side to download each track, and make sure you preview them through a DAC or soundcard capable of reproducing 24-bit/96Khz.
Get the tracks here (http://soundcloud.com/linnrecords/sets/24bit-comparison-alyn-cosker)
Follow us on Twitter (http://www.twitter.com/linnrecords) too if you would like to keep up with it and hear some more. Coming up in the series we have some classical recordings (at 24/192kHz!) some great rock and pop from Maeve O'Boyle, and as yet unreleased new work from jazz great Claire Martin!
Enjoy!
Jim - Linn Records
This is not a copyright violation is it?
I guess not if the a guy from the label is posting them.
It would be better if the tracks were downloadable, though (or am I just not getting how to do this?). Else you never know wether an audible difference is just due to improper quantization/dithering. I would like to be able to do that myself, before coming to false conclusions.
Can you make that possible, Jim?
Edit: Downloading is possible.
Down arrow on the right side will download.
terry
Hi Folks,
Yeah the small arrow on the right is a little subtle! Playing them online will give you the chance to contrast the downloaded files to a 128k MP3 stream - which is the what a lot of people listen to regularly, and in some cases exclusively! Once you have heard the beauty of the 24-bit Studio Master files its hard to go back to anything else.
Some really impressive recordings to come... so keep an eye out.
(and of course we are the copyright owners, so it's all completely guilt free!)
Jim
Thanks for making this available! I'll happily give it a try.
It's pretty courageous to make this public. Many people might not be able to hear a difference in a proper ABX setup. But if at least some do, you are going to win the whole pot.
Take a very long look at the waveform nulls before trying to ABX anything - IIRC, I was unable to null the samples to under perhaps -30dBFS or so. Some waveforms are visually different well above quantization noise. I'm not sure yet if this is some meaningless artifact of the conversion or if it represents a substantial DSP difference, but the sheer magnitude of the waveform differences does not leave me with a swell feeling.
EDIT: See below.
Let me first say this: This 24/96 track from Linn is an excellent recording, certainly well worth a buy!
But what most folks at Hydrogenaudio.org are interested in is mainly the question wether the 24/96
storage format is able to make an audible difference vs. Redbook, when all rules of proper mastering have been followed.
As the samples have been provided by Linn, objective comparison is not possible, mainly due to clipping
(and maybe also because of aspects Axon has touched). When you set your output sample rate to 96kHz Linn's Redbook sample cannot be SRC'ed at high quality without clipping. This will produce audible differences that are
not caused by limitations of the 16/44.1 storage format.
In the following I will describe a procedure, that you can employ to produce your own highest quality Redbook masters from Linn's HiRez sample and prepare them for ABX testing.
1. Convert the Flacs to WAV.
2. Download the excellent Sox from sox.sourceforge.net (http://sox.sourceforge.net/) and place the binary in the same folder as the Bheki WAVs.
3. Generate your own Redbook master from Linn's HiRez WAV (VHQ SRC with intermediate phase low-pass, noise shaped dither, 1db gain reduction to prevent clipping):
./sox Bheki.wav -b 16 Bheki-16bit-custom.wav gain -1 rate -v -I 44100 dither -s
4. Take the down-converted Redbook sample and create an upsampled 96kHz version from it. This eliminates any side effects for the later ABX comparison without increasing the actual resolution of the content. Again VHQ intermediate phase SRC:
Edit: Fixed wrong copy & paste:./sox Bheki-16bit-custom.wav -b 24 Bheki-16bit-custom-abx.wav rate -v -I 96000
5. Apply the same -1db gain to the HiRez sample, that you have applied to the Redbook sample to prevent clipping, without touching anything else:
./sox Bheki.wav -b 24 Bheki-abx.wav gain -1
Now ABX the HiRez
Bheki-abx.wav vs. the LowRez
Bheki-16-bit-custom-abx.wav and check if you can hear a difference.
If you're interested in how Linn's 16-bit compares to the HiRez version without the effect of clipping, try the following:
./sox Bheki-16bit.wav -b 24 Bheki-16bit-abx.wav gain -1 rate -v -I 96000
Then ABX the HiRez
Bheki-abx.wav. vs. Linn's LowRez
Bheki-16-bit-abx.wav.
As a last step ABX Linn's provided 16 bit mixdown vs. their HiRez sample.
I would provide these samples myself but that obviously
would be copyright violation.
Here's another publicly-available pair of clips , offered by the engineer for comparison of 16/44 to 24/96
http://www.soundkeeperrecordings.com/format.htm (http://www.soundkeeperrecordings.com/format.htm)
Maybe my setup, environment, ears (or a combination of all) aren't all that good because I hear no discerning differences by any of the files offered for comparision in this topic. I didn't even bother doing a blind test...I heard no differences knowing which track was which and would fail an abx easily.
Am I missing something or is there no real difference in the files? Perhaps someone could help me, or I'm one of the believers that 16-bit, 44khz is good enough for any equipment I'll ever own.
BTW, Linn's samples are of different length and not time-aligned. One should correct that before comparing with an instant-switching capable setup (as Foobar's ABX component). I'll start my round of testing now.
@Kitsuned: Don't worry. I expect few if any will be able to distinguish these in a proper ABX test.
Man, that's hard. I could swear that the low resolution sample sounded muffled compared to the high resolution version, when playing them separately in iTunes! No joke. But loaded into the ABX comparator the differences seem to be gone. I'm at 50/50 after 4 tries and now my ears ring due to the relatively high volume.
I'll continue as soon as the fatigue is gone.
I love Linn records, and Claire Martin especially.
Thanks for the free music!
Cheers,
David.
False alarm on my front - the nulling problems I was describing seem to have gone away entirely after upsampling to 384khz, attenuating -6db, sample-aligning, then downsampling back to 44.1khz. Comparing Bheki-16bit to Bheki after all that results in a frequency response variation of only +- 0.003db from 0-15khz and only +- 0.015db above that.
Note, however, that the 16-bit version appears to be attenuated by 0.165db relative to the high res version. This is just barely on the cusp of what I might consider audible.
I give it up, over 80% probability that I'm guessing after 10 trials. And I really did give it a chance: loud volume, several ranges tested, long runs. And the equipment wasn't bad either, a Benchmark DAC1 and a pair of Westone UM2, which tend to be my most revealing headphones.
I just compared the high resolution track against my own 16 bit version, since Linn's is not aligned and there's not much (probably nothing) you can do better than the above procedure. So after all that I don't buy the 24/96 advantage!
Still this is a nicely done record. They have used some kind of tube effect, that adds nice distortion to the louder amplitudes as some piano highlights, which makes it very pleasant to hear. Stereo image and setup are also well done. It didn't have to be mastered that loud considering it is a 24 bit record, but where else do you get that nowadays?
I tried ABXing the two files and failed. (MAudio Audiophile 24/96 > Arcam A80 > HD600)
Very nice sounding music though, thanks JimC. Something for me to investigate further.
For some reason the 24-bit one sounds like it has a more in-depth stereo feel to it than the 16-bit one. Now this is only noticed when listening at the original sound level, the bass seems to have more clarity too.
Maybe it is because you are listening with full knowledge of which is which?
I am sure ABX will make those differences go away.
In another HA thread, it has been found that for a given sample rate, dithering a 24-bit original with appropriate noise shaped dither to 16 bits makes no audible difference, for music recorded and played back at normal listening levels. (See http://www.hydrogenaudio.org/forums/index....st&p=626692 (http://www.hydrogenaudio.org/forums/index.php?s=&showtopic=49843&view=findpost&p=626692) )
It is very hard to make comparisons of 96KHz and 48KHz as a soundcard may operate slightly differently when called upon to process a different sample rate. Such differences as may exist for a human ear, seem likely to be extremely subtle. As a workaround, I note rpp3po's suggestion:
[blockquote]
4. Take the down-converted Redbook sample and create an upsampled 96kHz version from it. This eliminates any side effects for the later ABX comparison without increasing the actual resolution of the content.
[/blockquote]
Tonight I had a quick informal listen and I found the 24-bit version did sound slightly better; but then I realised the 24-bit version sounded slightly
louder. As Axon has mentioned, 0.165db will be barely on the cusp of what could be considered audible:
Note, however, that the 16-bit version appears to be attenuated by 0.165db relative to the high res version. This is just barely on the cusp of what I might consider audible.
Some people could detect it consciously. Others would not be able to detect it at all.
It is well known empirical result that a sound file that is slightly louder will seem slightly "better", all other things being equal.
I give it up, over 80% probability that I'm guessing after 10 trials. And I really did give it a chance: loud volume, several ranges tested, long runs. And the equipment wasn't bad either, a Benchmark DAC1 and a pair of Westone UM2, which tend to be my most revealing headphones.
I just compared the high resolution track against my own 16 bit version, since Linn's is not aligned and there's not much (probably nothing) you can do better than the above procedure. So after all that I don't buy the 24/96 advantage!
Still this is a nicely done record. They have used some kind of tube effect, that adds nice distortion to the louder amplitudes as some piano highlights, which makes it very pleasant to hear. Stereo image and setup are also well done. It didn't have to be mastered that loud considering it is a 24 bit record, but where else do you get that nowadays?
I have not gone to the trouble of preparing files in the manner rpp3po has described, and listening to them intently. Has anyone else gone through such an exercise? If so, what was the ABX result?
Common guys, I have just one pair of ears. Anyone else? So many people swear by 24/96. Convince me that it isn't just marketing hype.
I did an ABX but I did it differently than before. I lowered the 24/96 for 0.07 dB in volume. That way it matched the volume of the 16/44.1 track. The latter track was a wee bit longer, compared to that the 24/96 track lacked a few samples at the beginning of the track - needless to say, that I added them by copy/paste until the tracks were perfectly aligned (with sample accuracy). I didn´t do any upsampling - that would have changed the results. My external interface very neatly switches the samplerates around without me knowing it (no clicks involved) and foobar adds a pause when every track (A, B, Y or X) is changed.
This is the result:
foo_abx 1.3.4 report
foobar2000 v0.9.6.7
2009/08/19 10:14:17
File A: C:\Users\Eunice\Desktop\Bheki-16bit.wav
File B: C:\Users\Eunice\Desktop\Bheki.wav
10:14:17 : Test started.
10:15:46 : 01/01 50.0%
10:16:38 : 01/02 75.0%
10:18:57 : 02/03 50.0%
10:19:57 : 03/04 31.3%
10:21:22 : 04/05 18.8%
10:21:41 : 05/06 10.9%
10:21:58 : 05/07 22.7%
10:23:48 : 06/08 14.5%
10:24:46 : 07/09 9.0%
10:26:30 : 08/10 5.5%
10:27:57 : 09/11 3.3%
10:28:10 : Test finished.
----------
Total: 9/11 (3.3%)
I actually had a hard time figuring out where the differences were. I finally settled with listening to transients, in particular, the percussion set which is more transparent with the 24/96 file. The 16/44.1 track seems to be a bit more compact & flat in the soundstage and offers less "bite" and "air" but the differences are very subtle - as they are always with 24/96 material.
EDIT: I forgot the equipment I used. E-MU 0202 USB with a Sennheiser HD-600 directly fed from it. foobar uses ASIO.
I didn´t do any upsampling - that would have changed the results. My external interface very neatly switches the samplerates around without me knowing it (no clicks involved) and foobar adds a pause when every track (A, B, Y or X) is changed.
As far as I know the E-MU 0202 USB does not switch sample rates automatically, but employs Windows' mediocre quality realtime-SRC to convert to the rate set in the E-MU USB Audio Control Panel. So what you are testing is not a purer comparison, but HiRez vs. Windows' simplistic realtime-SRC. That alone would already be a comprehensible source of difference, but on top of that you get the problem I have described above. The 16 bit track, as it comes from Linn, cannot be converted to a higher sample rate (wether realtime or HQ) without resulting in several hundred clipped samples, so it must be attenuated prior to conversion.
Attenuation (of a 16 bit track) in addition requires re-dithering, which again raises the noise floor, so the approach to generate your own high quality master from the 24/96 source, with the best possible SRC and (only once applied) shaped dither, is the cleanest path to follow.
Common guys, I have just one pair of ears. Anyone else? So many people swear by 24/96. Convince me that it isn't just marketing hype.
Starting with the 24/96 file, lossyWAV
keeps 9.2 bits on average (default settings - generally transparent).
I looked at a specific section, which did not include any silence, and the most it kept was 13 bits, while the least it kept was 7 bits.
Compared with the Linn releases I have from a decade or more ago, that's a very "loud" recording with a lot of DRC and a lot of samples near digital full scale. I know it's not smashed to pieces 21st century pop, but it's
not a pure / audiophile / minimalist recording letting you hear the true dynamics of the music or instruments.
If you're going to ABX 24/96 vs 16/44, I think you've got to generate both yourself, like this.
1. use the provided 24/96 as a source (call that A)
2. halve the volume (23 bits! ) (call that B)
3. resample (B) to 16/44 (call that C) using fb2k or whatever
4. ABX (B) vs ©
5. resample © back to 24/96 (call that D)
6. ABX (B) vs (D)
None of these are the provided files - but the provided files are unsuitable for ABX because of the level difference, and the possibility of clipping in any standard oversampled DAC. (In my simulation, it's inevitable that there will be clipping at least once, unless there's headroom above 0dB FS in the oversampling process).
If you can pass 6, then there's a real difference when replayed in your system. If you can pass 4 but not 6, then there's a difference in the performance of your sound card / DAC, but it's only due to the limitations of your DAC, not fundamental to the difference between 44/16 and 24/96.
Cheers,
David.
As far as I know the E-MU 0202 USB does not switch sample rates automatically, but employs Windows' mediocre quality realtime-SRC to convert to the rate set in the E-MU USB Audio Control Panel.
Oh, but it does. Completely automatic. One of the few USB interfaces that can do so (the 0404 USB does the same). Remember, the 0202 is one of the few asynchronous interfaces out there with complete control over the USB connection. Just look at the manual, you can download it for free from E-MU. It also offers a completely useless tool called "E-MU USB Audio Application" where it shows the actual samplerate and USB connection used. I did not use it in this test since I would have been able to tell which track was which without even listening. I never use it all - saving resources.
The 16 bit track, as it comes from Linn, cannot be converted to a higher sample rate (wether realtime or HQ) without resulting in several hundred clipped samples, so it must be attenuated prior to conversion. Attenuation again requires re-dithering, which again raises the noise floor, so the approach to generate your own high quality master from the 24/96 source is the cleanest path to follow.
Exactly. That´s why I used the 24/96 track for editing and left the 16/44.1 completely unspoiled. The quantization noise resulting from editing the 24/96 file should be at around -121 dB if I´m not mistaken - and I did not use dithering when saving the edited and volume attentuated 24/96 track.
1. use the provided 24/96 as a source (call that A)
2. halve the volume (23 bits! ) (call that B)
3. resample (B) to 16/44 (call that C) using fb2k or whatever
4. ABX (B) vs ©
5. resample © back to 24/96 (call that D)
6. ABX (B) vs (D)
What's the point of repeating above procedure with the difference of halving the volume (-1db would be sufficient) and omitting dither? Both aspects are additional source of error.
Just look at the manual, you can download it for free from E-MU.
I did (http://www.emu.com/support/files/storage/0202-USB-EN.pdf), that was the source of my claim. Page 12, about the E-MU USB Audio Control Panel: "5. Sample Rate - Allows you to set the system sample rate...".
It doesn't talk anywhere about being able to switch automatically.
Just look at the manual, you can download it for free from E-MU.
I did (http://www.emu.com/support/files/storage/0202-USB-EN.pdf), that was the source of my claim. Page 12, about the E-MU USB Audio Control Panel "5. Sample Rate - Allows you to set the system sample rate...".
It doesn't talk anywhere about being able to switch automatically.
True. That is written there. I´m sorry for referring to the manual. They don´t even talk about automatic switching! How could they forget it? Well, with an actual application it looks like this: "Allows you to set the sample-rate" - but only when no data is being sent to the interface. With ASIO it switches automatically. I´m very sure that other owners of the 0202 could clear that up.
The ASIO protocol knows commands that allow an application to request a specific sample rate. So if Foobar is doing this, what you are saying about the 0202 makes sense.
1. use the provided 24/96 as a source (call that A)
2. halve the volume (23 bits! ) (call that B)
3. resample (B) to 16/44 (call that C) using fb2k or whatever
4. ABX (B) vs ©
5. resample © back to 24/96 (call that D)
6. ABX (B) vs (D)
What's the point of repeating above procedure with the difference of halving the volume (-1db would be sufficient) and omitting dither? Both aspects are additional source of error.
Whether you dither or not is up to you - I didn't say you should or you shouldn't - it's irrelevant on the 24-bit source, and it's just one of several choices you
have to make if you specify the 24/96 > 16/44 conversion parameters yourself.
Why not -1dB? Mainly because halving the volume is a very "clean" mathematical operation, with or (especially) without dither. -1dB is probably fine too, but leaves the possibility of mathematical errors in the scaling algorithm, specific systematic errors (effectively non-linear quantisation steps) if dither isn't used, and the possibility of driving certain trashier sound cards (admittedly ones completely unsuited to this test) into slightly non-linear regions.
Given that we're chasing a small (non existant?) difference, better to be safe than sorry.
...but my post was in response to Cavaille's procedure, not yours.
Cheers,
David.
P.S. In case you're about to, please don't critique these suggestions as if they come from a clueless audiophile - I've already implied what I think - i.e. ~14 bits is probably sufficient here. Even so, it's still worth doing the test properly.
(Even if the material, as I've already said, hasn't preserved the dynamics of a live performance, and uses a disappointingly high amount of DRC for Linn records - never mind for 24-bits!).
David, maybe my question got across a little rough. I don't question your expertise in digital audio.
When there's only ~14 bits above noise, I'm wondering, what Cavaille is hearing. There's no problem conserving this within the specs of Redbook. But maybe it's HF content, for which I have yet to find a paper confirming audibility in a controlled environment. Industry interest would be great, suspicious when nothing gets found anyway.
Not to put tooooo fine a point on this, but Cavaille is, statistically, predicted to have a much higher hearing limit than (most of) the rest of us.
That said, I really would like to see ABX results with 2BDecided's test.
Guys, I decided not to do an upsample of the the 16 Bit material because it would have colourized the outcome when compared to the original 24/96 (and yes, it would have introduced several distortions from clipping - just like rpp3po said). I think that we all know that every resampler changes the sound (even if only slightly) and I also think that we all know that differences between 24/96 & 16/44.1 are subtle at best. By upsampling I would have added another slight subtlety - that way the outcome would not be representable.
I simply turned to editing the 24/96 file in order to have higher headroom. BTW, even with that I wouldn´t have heard quantization errors - simply because the material is too loud for this and quite a bit compressed with brickwall limiting.
2BDecided´s way of doing a test is complicated but seems to be the best method so far.
And why should I (statistically) have a higher hearing limit than others here? I´m almost 34 years old and my hearing goes up to 18.500 Hz - perfectly normal, I´d say.
2. halve the volume (23 bits!) (call that B)
If you save this as float32, you can retain all 24 bits (for the subsequent conversion at least).
Why not -1dB? Mainly because halving the volume is a very "clean" mathematical operation, with or (especially) without dither. -1dB is probably fine too, but leaves the possibility of mathematical errors in the scaling algorithm, specific systematic errors (effectively non-linear quantisation steps)
BTW, isn't the potential quantization error when applying a -1db attenuation (much) smaller than 6db in any thinkable case?
Halving of the volume wouldn't add any mathematical error <6db but raise the noise floor by +6db instead, since you just blank the LSB. Then what would be the advantage of adding 6db uncorrelated noise in your proposal over adding <6db uncorrelated noise (the distribution up and down rounding errors should be uniform) in my proposal?
On the bottom line this is just nit-picking, though. There should be
plenty of headroom left with both approaches. Also I may overlooked something, and then I'd be interested in the explanation.
So has anyone looked at /listened to the *other* two files, that I linked to, at Barry Diament's site?
They are mastered very 'old school' -- just a single true 'peak' sample per channel. Crest factor ~20 dB. Less than .1 dB difference between corresponding channels. Spectral content up to the limits of each format.
I have downloaded them: 36MB and 118MB. In a quick informal test the two versions seem to sound slightly different on my system in terms of their frequency response, which may be a comment on my system! Haven't had the chance to analyse the files.
So has anyone looked at /listened to the *other* two files, that I linked to, at Barry Diament's site?
I hadn't, because I assumed you were linking to something I'd already seen, but these are new.
They're great recordings. I can't ABX them, but I found something interesting (well, if your "interesting" threshold is really low and skewed towards obscure audio issues!).
In fb2k 0.9.6.2, WinXP, M-audio 2496, the files play at a slightly different pitch/speed! The 44.1 / 16-bit version is fractionally too fast / high pitch.
There's some sample rate conversion going on, because I know the 2496 mutes slightly when it changes sample rate, whereas it
doesn't mute when I switch from one file to the other. However, I don't have fb2k's sample rate conversion enabled - in fact, when I drop SSRC into the DSP chain to convert everything to 96kHz, the problem goes aware.
Very strange.
If I try to ABX them, the problem goes away
and the 2496 mutes between the tracks as it changes sample rate - so the playback route is different within the ABX comparator than in normal playback.
I thought I was going mad and imagining it - but I've just loaded up 20 copies of the tracks in the play list, randomised them, and with my eyes closed I can pick out the 44.1kHz ones, because of the slightly higher pitch.
I don't know if this is some configuration freak on my PC, but it shows the kinds of problems you can face in such a comparison.
I've just downloaded the free mp3 extracts - it sounds like a really nice CD. Not sure about the comment on their website about the difference between CDs and CD-Rs (they charge more for a CD-R!), and the fade-in of the 24-bit file is only 16-bits (very strange!) and there's some bit-sticking during the fade out too (have a look - continuous runs of identical samples - can't happen by accident!), but the music recording seems fine.
Cheers,
David.
David, I used Audacity at a project rate of 96KHz. That allowed me import the 24/96 file first and then the 16/44 file. I was the able to time align the two files despite their different sample rates.Here's another publicly-available pair of clips , offered by the engineer for comparison of 16/44 to 24/96
http://www.soundkeeperrecordings.com/format.htm (http://www.soundkeeperrecordings.com/format.htm)
The 24/96 file appears to be 2 or 3 samples (at 96KHz) ahead of the 16/44 file, at the beginning of the music. When time aligned, and with one file inverted, the result is inaudible at a moderate listening level (at least at the beginning; there seems to be a very slight drift in the alignment later in the files). A spectrum analysis shows some low level content below about 400Hz, and some high frequency content commencing at about 16KHz; but the null is quite good.
I imagine though that many people might prefer to downsample the 24/96 file to 16/44.1 and upsample that to 24/96, for critical ABX testing.
Yes - same in CEP - I can resample one to the other can get alignment within a fraction of a sample, but not perfect. To be expected with different resampling algorithms - there's nothing devious going on.
Cheers,
David.
I've just downloaded the free mp3 extracts - it sounds like a really nice CD. Not sure about the comment on their website about the difference between CDs and CD-Rs (they charge more for a CD-R!), and the fade-in of the 24-bit file is only 16-bits (very strange!) and there's some bit-sticking during the fade out too (have a look - continuous runs of identical samples - can't happen by accident!), but the music recording seems fine.
Cheers,
David.
Re: comments on the site, Diament adheres to a significant amount of audio woo, as witnessed by his webpage on the benefits of vibration control, and many dubious claims in his posts to Steve Hoffman's site -- but he claims to be a 'purist' in these matters which is why I thought these samples would be ideal for our purpose here. It should be possible to ask him directly why you're seeing the things you're seeing (like the 16-bit fades)
I gave it a try, yet I couldn't ABX them. My score after 6 runs was 3/6, which is exactly 50%, and what you get when totally guessing without any clue. I had the subjective impression that the instruments sounded more voluminous with the 24/96 file, yet I failed utterly to ABX them.
- I used the 24/96 source file to create my own master
- opened it with Audition
- normalized to 93% to prevent clipping
- saved normalized file as "Bheki 2496.wav"
- than downsampled (including dithering) the still loaded normalized file to 16/44 and saved it as "Bheki 1644.wav"
footbar2000 Asio > Asus Xonar dx > HD 555
I am sorry Linn Records, but I am unable to obtain persuasive ABX results, by taking the 24/96 version, reducing it by 1dB, downsampling that to 16/44.1 with noise-shaped dither (Audacity software), upsampling back to 24/96 and comparing with the 1dB reduced version of the original.
This is both with your track referred to at post #1, and with the SoundKeeper Recordings track referred to by krabapple at post #9.
With the two tracks provided (a jazz instrumental piece, and a male vocal with guitar accompaniment), using a reasonable quality sound reproduction system, any audible difference that exists appears to be so subtle as not to be detectable with average adult human hearing.
I imagine that if some adults can detect a difference, it will be a very minor difference for those people, and perhaps noticeable only in direct A B comparisons.
I note that Cavaille's positive ABX result (though he went to the trouble of correcting for the identified amplitude disparity between the files provided for comparison) was obtained without the rigour of presenting his DAC with the same sample rate. DACs use different filter implementations at different sample rates, and there are many choices available for implementing a filter for a 44.1KHz sample rate (as there are trade-offs involved as between upper frequency response and avoiding phase artifacts). It is my understanding that these choices are likely to be greater in their potential effect on the audible sound than the effect of an upsampling from 16/44.1 to 24/96.
3. Generate your own Redbook master from Linn's HiRez WAV (VHQ SRC with intermediate phase low-pass, noise shaped dither, 1db gain reduction to prevent clipping):
Hi rpp3po, I tried sox with the intermediate phase setting, but when I subtracted the resulting file from the original (the 24/96 files less 1dB), there was quite noticeable (to my ears) higher frequency sound. So I ended up using Audacity, with dither set to noise-shaped. This gave a very good [for my ears] null.
I still have an open mind on the 96KHz vs 44.1KHz question, but the 24/96 tracks provided do not appear to be audibly different to versions of them downsampled to 16/44.1 and then upsampled back to 24/96.
3. Generate your own Redbook master from Linn's HiRez WAV (VHQ SRC with intermediate phase low-pass, noise shaped dither, 1db gain reduction to prevent clipping):
Hi rpp3po, I tried sox with the intermediate phase setting, but when I subtracted the resulting file from the original (the 24/96 files less 1dB), there was quite noticeable (to my ears) higher frequency sound.
In general, subtracting one audio signal from another is not very enlightening, esp. where perceptual techniques (e.g. intermediate phase filtering) have been used.
Using Audacity should be fine, as the noise shaper and SRC implementations are very good. The applicability of auditioning a difference file to assess the perceptibility of an effect/conversion is debatable, as it doesn't take the effects of masking into account.
My choice of an intermediate phase filter adds more post-ringing noise than, for example, a linear phase filter, but should nevertheless be less audible (because more pre-ringing is eliminated). This applies to the signal itself, but not necessarily to the difference file.
But as said, your conversion method is fine, and well suitable to produce a conclusive results for your ears.
Edit: Bandpass was faster.
After reading this post (#16) (http://forums.linn.co.uk/bb/showthread.php?tid=3663&pid=35927#pid35927"%20target="_blank">http://forums.linn.co.uk/bb/showthread.php...=35927#pid35927) I would very much like to see JimC perform a level matched ABX test.
Are there even people who are able to do that? And if there are, what do you have to lock, or rather, listen for?
After reading this post (#16) (http://forums.linn.co.uk/bb/showthread.php?tid=3663&pid=35927#pid35927"%20target="_blank">http://forums.linn.co.uk/bb/showthread.php...=35927#pid35927) I would very much like to see JimC perform a level matched ABX test.
JimC seems to be a marketing employee at Linn. From that perspective how something is
perceived is more important than what actually
is. So don't expect him to care too much. From many audiophile (and their own) forums he is probably used to people who really
want to belief that 24/96 is superior. And he just feeds them, what they want to be fed with.
I wonder how he tapped into HA, inviting to evaluate the superiority of a HiRez track with roughly only 90 db dynamic range. He probably expected this thread to evolve differently.
I wonder how he tapped into HA, inviting to evaluate the superiority of a HiRez track with roughly only 90 db dynamic range. He probably expected this thread to evolve differently.
Doubt it - either he doesn't know what HA is for at all and is just spamming every audio forum on the internet (that's a marketing person's job), or he does, but think some people here will still be interested in nice quality recordings, whatever the sample rate / bit depth.
I ordered the http://www.soundkeeperrecordings.com/ (http://www.soundkeeperrecordings.com/) disc - the music was quite nice too, which is more than you can say for many "audiophile" recordings.
btw, I also stumbled on...
http://www.hdtracks.com/ (http://www.hdtracks.com/)
...which has lots of 24/96 recordings - and (more importantly for me) lossless downloads of decent quality recordings. Not cheap though! I think I'm decidedly old fashioned in my love of physical media - a FLAC download is more convenient than a DVD-V with audio, but it doesn't feel like you're buying as much!
Cheers,
David.
Let me first say this: This 24/96 track from Linn is an excellent recording, certainly well worth a buy!
I trust you're just being nice, because the dynamics of this recording have been compressed with a pretty heavy hand.
I couldn't find any musically-related dynamics much in excess of 30 dB.
If you discount the obviously artificial fade-out, it has minimal actual musically-related dynamic range.
If someone told me that it was a 16/44 recording upsampled to 24/96 and backfilled above 16 Khz with with shaped l noise, I'd be hard put to disprove that hypothesis.
btw, I also stumbled on...
http://www.hdtracks.com/ (http://www.hdtracks.com/)
...which has lots of 24/96 recordings - and (more importantly for me) lossless downloads of decent quality recordings. Not cheap though! I think I'm decidedly old fashioned in my love of physical media - a FLAC download is more convenient than a DVD-V with audio, but it doesn't feel like you're buying as much!
The "Dragon Boats" sample is not nearly as compressed as the Linn sampler. It might even have natural dynamics or just light compression. Still no *serious* dynamics that would tax a 16 bit system, other than one more boring artificial fade-out.
Has no one heard of recording reverb tails as an exercise in demonstrating natural dynamic range?
I think they have, but they've learned that natural reverb gets lost in room noise long before it taxes 16 bit resolution.
It's supposedly a direct-to-stereo acoustically / naturally mixed recording.
There's obviously at least a gain change - you'd never record that close to 0dB FS live and be lucky enough not to clip!
Cheers,
David.
I am sorry, but it appears to me that many of you don't know what you are
doing here.
When starting from a 96kHz master it is impossible to compare 96kHz with 44.1kHz.
What you are comparing is the 96kHz source with [[[ the combination of a 44.1kHz sampled signal space with the particular anti-alias filter used when converting from 96kHz to 44.1kHz ]]].
So before starting you have to agree on which AA filter you'll include.
"Generate your own Redbook master from Linn's HiRez WAV (VHQ SRC with intermediate phase low-pass"
SoX with intermediate phase is, obviously, non-linear-phase and thus will add phase distortion in the 10-22kHz band of the 44.1kHz version. You introduce a difference before you even started ... (hence the failing nulls).
Maybe that is what you wanted to test, but I doubt it.
What AA filter to use then during downsampling? (edit: typo)
One possible choice would be a linear phase half-band filter with not too much of transition band attenuation, mimicking the type of filter that is part of the typical recording ADC.
It injects pre- and post-ringing at 22.05kHz and it has some aliasing. Upon replay such a signal will force the typical DAC to pre-ring too, on top of the AA's ringing.
Another possible choice would be a linear phase filter with cut-off slightly below 22kHz, say 21kHz, and a very steep transition band rolloff. It injects pre- and post-ringing at 21kHz, but it has no aliasing. Upon replay such a signal will not force the typical DAC to pre-ring too. There's only the AA's ringing.
The minimum phase and intermediate phase AA filters exist to get rid of the pre-ringing, at the cost of phase distortion. Whether upon replay there still is pre-ringing depends entirely on the particular configuration of the AA filter and the choice of replay filter in the DAC (or software oversampler).
"I didn´t do any upsampling - that would have changed the results. "
Quite the contrary, in fact. Not doing oversampling in a controlled way in the software, prior to sending the 44.1kHz to the DAC leaves all oversampling, including the reconstruction filter, at the mercy of the DAC. And typical DAC oversampling/reconstruction filters are linear-phase half-band, i.e. with pre- and post-ringing at 22.05kHz and imperfect stop band rejection.
Such a reconstructor violates Shannon's theorem, which prescribes the Sinc(t) function for replay. So, in order to do a valid test, take the DAC out of the equation and use SoX for oversampling, with settings that approximate a Sinc function (linear phase, highest quality, highest steepness, 22kHz or slightly less).
"I didn´t do any upsampling - that would have changed the results. "
Quite the contrary, in fact. Not doing oversampling in a controlled way in the software, prior to sending the 44.1kHz to the DAC leaves all oversampling, including the reconstruction filter, at the mercy of the DAC. And typical DAC oversampling/reconstruction filters are linear-phase half-band, i.e. with pre- and post-ringing at 22.05kHz and imperfect stop band rejection. ...
Thanks WernerO for your various observations.
I think we are left with the fact that if we are to attempt any comparisons between 24/96 and 16/44.1, extreme care is needed in creating and playing the 16/44.1 files. An interesting article I came across today (which many HA members may have already seen) on the topic of ant-aliasing filtering for 44.1KHz,
Ringing False: Digital Audio's Ubiquitous Filter, by Keith Howard in stereophile January 2006, can be found here (http://stereophile.com/reference/106ringing/index.html).
If there are any audible differences, they appear to be likely to be extremely small.
Howard is a white hat but alas for the usual Stereophile pandering to subjectivism, placing its 'findings' on par with those of formal tests:
listening tests were undertaken to establish formally that the anti-alias filtering required for a nominally flat response to 20kHz, even at the 44.1kHz sampling rate—the sternest test because the transition band from 20kHz to 22.05kHz is so narrow, demanding an extremely steep filter rolloff—has no audible effect. Many informal listening tests were conducted too, often using Sony's PCM-F1, because it was the first 16-bit digital recorder that most people were able to lay hands on. The outcomes of this testing, formal and informal, were overwhelmingly positive. Many PCM-F1 users claimed that a signal passed through the machine's A-to-D and D-to-A stages was indistinguishable from the feed, and some still cite that experience as proof that 16-bit/48kHz audio, properly realized, is essentially perfect.
But even in the earliest days of domestic digital audio there were dissenting voices. Many hi-fi writers, myself included, were thoroughly underwhelmed by our initial experiences of Compact Disc, and so were some influential audio professionals, such as Doug Sax*. Over a period of some years the intensity of this opposition to CD decreased somewhat, but many commentators and ordinary audio consumers concluded that there was something fundamentally amiss with 16/44.1 and 16/48 audio. Many of them voted with their feet, continuing to prefer the sound of the "obsolescent" LP.
AND, the 'various experiments' he cites to support response to >20kHz playback, are all from the unreplicated work of Oohashi
et al. *who uses Shakti Stones.
I am sorry, but it appears to me that many of you don't know what you are
doing here.
When starting from a 96kHz master it is impossible to compare 96kHz with 44.1kHz.
What you are comparing is the 96kHz source with [[[ the combination of a 44.1kHz sampled signal space with the particular anti-alias filter used when converting from 96kHz to 44.1kHz ]]].
So before starting you have to agree on which AA filter you'll include.
"Generate your own Redbook master from Linn's HiRez WAV (VHQ SRC with intermediate phase low-pass"
This sounds like a red herring argument to me. No matter what wider-bandpass source you start out with including a wide-bandwidth analog source, you're converting from a wider-bandwidth source to a lower-bandwidth source.
Before one looses too much sleep over this, they ought to try the experiment with whatever down-conversion product they have, and see if they get a positive result. Then if you get positive results. we can do a post-mortum.
Back in 2002 or so I posted a goodly number of downsampled files, made from 24/96 recordings that I made with a wide-bandwidth production system including mics that were reasonably flat past 40 KHz. These recordings were arranged so that they naturally containe exceptional amounts of energy > 20 KHz. I did the resampling with CoolEdit Pro. Nver heard any postive results involving anything but some training files that involved downsampling to some rediculous sampling rate like 28 KHz.
they ought to try the experiment with whatever down-conversion product they have, and see if they get a positive result. Then if you get positive results. we can do a post-mortum.
According to this HA thread (http://www.hydrogenaudio.org/forums/index.php?showtopic=74651) it might be time for a post-mortem. There seems to be a positive ABX for a 24/96 to 16/44.1 conversion.
they ought to try the experiment with whatever down-conversion product they have, and see if they get a positive result. Then if you get positive results. we can do a post-mortum.
According to this HA thread (http://www.hydrogenaudio.org/forums/index.php?showtopic=74651) it might be time for a post-mortem. There seems to be a positive ABX for a 24/96 to 16/44.1 conversion.
So you didn't note or follow up on the coment in the heading: "Successful ABX of TPDF white dither vs. noise-shaping at normal listen, Editorial note: Seems there are other issues besides dither at play."
Seems there are other issues besides dither at play."
Sure, and it's a pity that the OP doesn't seem to see the importance of separating the issues of SRC and dithering/re-quantization. However, if he can hear a difference between a 24/96 and a 16/44.1 source, that deserves some attention IMO since according to HA belief this should not be possible unless something has been done wrong.
Seems there are other issues besides dither at play."
Sure, and it's a pity that the OP doesn't seem to see the importance of separating the issues of SRC and dithering/re-quantization. However, if he can hear a difference between a 24/96 and a 16/44.1 source, that deserves some attention IMO since according to HA belief this should not be possible unless something has been done wrong.
Hence the interest in the origional commerically-available track.
True scientific skepticism does not deny that the possibility that there may be some audio some place that actually sounds different when downsampled right.
Our present findings are due to us not having the good fortune of actually finding that audio, given many energetic efforts to find it. While we might suspect that such a piece of audio does not exist, we have to remain open to the possibility that it may exist.
There is also the possiblity that some errors in the prepartion of the test may explain its results. One way to convince ourselves of the adquacy of the test results that have been reported is for us to replicate them ourselves, starting with the origional source material.
....this should not be possible unless something has been done wrong.
Until the original is provided and processed in the prescribed manner we would be foolish to make
any assumption regarding the correctness of the processing of the dithered samples that have been provided.
As far as I know the E-MU 0202 USB does not switch sample rates automatically, but employs Windows' mediocre quality realtime-SRC to convert to the rate set in the E-MU USB Audio Control Panel.
Oh, but it does. Completely automatic.
The following comment by Cavaille in another [a href='index.php?showtopic=75277']thread[/a] made me look at our only claimed positive 24/96 vs. 16/44.1 ABX result in this thread in a different light:
I use the 0202 USB [...] Both interfaces also have one big problem: they don´t work very well with foobar´s ASIO output. It has to be configured anew after every restart of the host-pc.
Looks like it might not have been as "automatic" as claimed and it might have been flawed switching after all and not golden ears that made the difference. I restate: prior upsampling using a high quality converter is a much cleaner approach to evaluate wether 16/44.1 discards audible content that 24/96 would preserve.
I use the 0202 USB [...] Both interfaces also have one big problem: they don´t work very well with foobar´s ASIO output. It has to be configured anew after every restart of the host-pc.
Looks like it might not have been as "automatic" as claimed and it might have been flawed switching after all and not golden ears that made the difference. I restate: prior upsampling using a high quality converter is a much cleaner approach to evaluate wether 16/44.1 discards audible content that 24/96 would preserve.
My friend, you try to punch holes into something where nothing is contradicting. I suspect you do that just so you can keep your opinions alive - whatever the costs and whatever they are.
So let´s read the following sentence very carefully: The E-MU 0202 USB switches sample rates automatically (it doesn´t have an internal resampler). BUT: when you re-start your PC you´ll have to configure foobar2000s ASIO output.
Let´s read again: foobar2000´s ASIO output. Not the interface itself.
Reason: foobar2000 somehow uses the hardware-ID for assigning the ASIO output (there is a thread for this somewhere on HA). Since the interface always gets a different Hardware-ID after every re-start due to the special synchronized USB connection, foobar2000 forgets all the information. So this is not a failure of the interface, it is a failure of... well, it should be obvious by now. Several users complained about this but apparently nothing will be done - and I don´t want to indulge about the reasons for this.
Every other ASIO-capable program I use (WaveLab, iZotope, Sound Forge, hell, even WinAmp (!)) works every time with the interface. It is just a problem of foobar2000.
I don´t know what you´re talking about here but I had to correct this.
I ordered the http://www.soundkeeperrecordings.com/ (http://www.soundkeeperrecordings.com/) disc - the music was quite nice too, which is more than you can say for many "audiophile" recordings.
As a follow up, a friend was saying that he thought mp3 players were just toys (though he had and used one), and he thought music was best from vinyl.
So I played him the Dragon Boats track from my Sansa Clip through HD580 headphones.
I can't say that I "converted" him there and then, but he was very impressed! He was amazed that you could really hear the space in which it was recorded.
Cheers,
David.
So let´s read the following sentence very carefully: The E-MU 0202 USB switches sample rates automatically (it doesn´t have an internal resampler). BUT: when you re-start your PC you´ll have to configure foobar2000s ASIO output. Let´s read again: foobar2000´s ASIO output. Not the interface itself.
Technically, that's not accurate. Since the 0202 doesn't employ an ASRC the sample rate has to be set in software (and ASIO has provisions to do that automatically). A hardware sample rate detection would mean a considerable amount of samples lost during rate detection.
And yes, I don't believe in your results (or at least conclusions). It is against anything known, that has been conducted in a controlled environment. There is no audible difference for 24/96 for humans at the listening levels we have had here. The only advantage of 96 kHz is that cheaper filters can accomplish what takes some effort to be inaudible at 44.1 kHz. If you live in the Stuttgart region, how about we repeat the exact same test at the Max Planck Institute in Tübingen? Only condition would be that a 2nd control test is conducted with pre-converted files (96->44.1->96 vs. 96) to rule out side effects of the 0202.
Reason: foobar2000 somehow uses the hardware-ID for assigning the ASIO output (there is a thread for this somewhere on HA). Since the interface always gets a different Hardware-ID after every re-start due to the special synchronized USB connection, foobar2000 forgets all the information. So this is not a failure of the interface, it is a failure of... well, it should be obvious by now.
And BTW, the 0202 probably doesn't use "synchronized" (isochronous) USB, that's about the worst mode there is for audio (common for budget gear), but asynchronous (isochronous) USB, which can be designed to have very high jitter tolerance. With asynchronous USB the audio device signals the exact rate at which the host should push audio frames. Since the audio hardware is in total control over the clock and the host just "listens & obeys", the sample rate must be known by the audio device in advance, it cannot be "detected". This is either done through a driver control panel or, in the case of ASIO, sent as side information of an audio stream to the driver, which extracts it and sends a command to the device to switch the pulling rate.
More thread fodder --
Diament has now posted three versions of a track, in 16/44, 24/96, and 24/192
http://www.soundkeeperrecordings.com/format.htm (http://www.soundkeeperrecordings.com/format.htm)
There are three new samples up on the Soundkeeper Recordings website:
http://www.soundkeeperrecordings.com/format.htm (http://www.soundkeeperrecordings.com/format.htm)
44.1kHz vs 96kHz vs 192kHz this time.
They're from a soon-to-be-released album. Beautifully recorded, just like the first. I wish music that I like was recorded with such care.
(I'm surprised to see a laptop present in the recording room itself).
Cheers,
David.
EDIT - DOH! I guess you're on his mailing list too then krab!
EDIT - DOH! I guess you're on his mailing list too then krab!
Nope...i have other informants.
One instrument on this new recording has "lots" of energy above 20kHz. Everything else would be basically untouched by a 20kHz low pass filter, even if humans could hear up to 40kHz.
Still, you've gotta love Stereophile (talking about the previous release) "...It's not that there's so much going on in those extended high frequencies, but it's obvious that they're not there at all on the "Red Book" CD. It's not so much a question of night and day as one of life and death..."
Now you'd think life vs death would be ABXable, wouldn't you? When confronted with an ABX test, can Stereophile journalists no longer tell whether they're dead or alive?
Cheers,
David.
Now you'd think life vs death would be ABXable, wouldn't you?
Well, do you really think you could ABX beeing dead?
Hm... maybe we are all going to hell for our outrageous scepticism and our punishment shall be to do 16/44.1 vs. 24/96 ABX testing for the rest of eternity.
Now you'd think life vs death would be ABXable, wouldn't you?
Well, do you really think you could ABX beeing dead?
You may concentrate on the sound of breathing, not breathing.
It appears this thread has sparked some interest over on Steve Hoffman's board.....
http://www.stevehoffman.tv/forums/showpost...p;postcount=606 (http://www.stevehoffman.tv/forums/showpost.php?p=5692196&postcount=606)
Start there and read ahead to savor this classic bit of audiophilery:
The Hydrogen guys are inexperienced. Anyone who has done even a modest amount of hirez recording understands the value of 24/96 over 16/44. It's simply not subtle, no matter how good the 16/44 playback is.
"
It seems to me we have two audiophiles - those that listen to music, seek out the best, and enjoy it and those that debate the scientific merits endlessly.
I don't want to shut down the technical debate if it is done in an open-minded way and subjective assessments are included...after all, you cannot measure everything we hear. Sadly I see little open-mindedness at Hydrogen.
It seems to me we have two audiophiles - those that listen to music, seek out the best, and enjoy it and those that debate the scientific merits endlessly.
Funny, but I thought there are two kind of "audiophiles" indeed. Those who enjoy the music on equipment sufficient for the task, and those who listen to their equipment and not to the music.
More of the usual "of course you can easily hear a difference, it's obvious" - which would, if pressed, magically turn into "I can't ABX a difference".
I like the honest post that says "the differences are subtle". I think that's the conclusion that any rational person must come to - if there are any audible differences, they must be very subtle, because positive test results are so rare.
Consider this...
1. People say "it's all about having revealing gear" - now, I can see why that could be true in theory.
2. People say "it's all about having golden ears" - again, I can see why that could be true in theory.*
But if both of these things were true in practice for 24/96, people selling "revealing" gear would do tests to prove that the 16/44 vs 24/96 difference was provably genuinely audible through their equipment (at least for the "golden ears" who could hear it).
How great a piece of marketing would this be? "Buy our speakers - you can really hear the benefits of 24/96 - here are the test results to prove it".
Such evidence and marketing is notable by its absence. Draw your own conclusions.
Cheers,
David.
P.S. * - number 2 is true in practice for many provably real, audible, measurable but subtle differences. Psychoacoustic test results (measuring the ability to hear the barely audible) prove this all the time - results are determined by 1) raw ability and 2) training. No one would argue with this.
More of the usual "of course you can easily hear a difference, it's obvious" - which would, if pressed, magically turn into "I can't ABX a difference".
I like the honest post that says "the differences are subtle". I think that's the conclusion that any rational person must come to - if there are any audible differences, they must be very subtle, because positive test results are so rare.
Consider this...
1. People say "it's all about having revealing gear" - now, I can see why that could be true in theory.
2. People say "it's all about having golden ears" - again, I can see why that could be true in theory.*
If you understand all of the theory, then you conclude that its not all about just any one thing.
If you want the most sensitive test possible, its about the source material, *and* the equipment, *and* the room, *and* the listener training, *and* the listening procedures. The best complete statement about all of that is AFAIK ITU Recommendation BS 1116.
Anybody who goes off on just one or two of the above is basically telling us about how incomplete his knowlege and experience are.
The post from the stevehoffman forum is almost insulting! Some people appear to be thinking that there´s a war going on between either side. Such unnecessary, time consuming bulls**t. However, as stupid as that comment is I always was under the impresson that "both sides" need each other: the fools for making stupid propositions and the realistic guys to prove them wrong or right. Just imagine, there could be coming something good from it after all. It doesn´t appear to be right now but maybe in the future... I also try to think up an example from the past - right now it slips my mind.
BTW, David could it be argued that higher samplerate is more important after all? This may be farfetched but I draw conclusions from your listening tests of brickwall limiting where I found the differences subtle but existing.
The point is indeed wether the differences are heard blind or not blind. I used to head differences between power cables and such kind of stuff. I can confirm that they were not subtle. They were obvious... and they were disappearing under blind listening conditions. Sometimes even just before the blind test begins.
For some people, the differences remain audible during the blind test, while their answers are no more exact than randomness. Superb examples can be dug out from the listeners comments in two power cable blind tests linked in the new topic : http://www.hydrogenaudio.org/forums/index....showtopic=82777 (http://www.hydrogenaudio.org/forums/index.php?showtopic=82777)
So yes, audible differences are big. Obvious. But they are not caused by the sound. They are in the listener's head. And that's no offense, as James Jhonston said in the presentation of Ethan Winer, "Audio Myths Workshop", at the AES in 2009.
If I play the same thing twice [...] you will remember two different things. This is no hallucination, this is no deception, [...] it's not those... pejorative word... it's just how human being works.
If you have reason to think things are different, you're likely to listen differently, you're likely to focus on different things, and you're likely to remember different things.
Or if you're maybe convinced everything sounds the same... you're still yourself that way.
So this makes the point whenever you do auditory testing, that any test that you do for auditory stimulii, it has to have falsifiable design. It means that you have to actually be able to tell if something is actually detected or if its something that you either conciously or subconciously steered yourself into notice. You have to be able to test that. Now, it basically means that you have to do a blind test.
Because if you know what the two things are, your brain is gonna use this information. I don't care how smart you are, how trained you are, who you are, there hasn't been examples yet of somebody who can avoid. It's just life.
People that hear big differences that way are a minority, but they exist. Many people hear subtle differences that way, and some not at all, I think. Or maybe they can hear small differences, but they don't state it because they know that their perception changes, and they directly translate it into "I don't hear any difference".
It's all about the origin of the difference, the ear-brain system or the audio source, rather than about the size of the difference.