• WANTED: Happy members who like to discuss audio and other topics related to our interest. Desire to learn and share knowledge of science required. There are many reviews of audio hardware and expert members to help answer your questions. Click here to have your audio equipment measured for free!

New paper on high resolution audio perception without music

Blumlein 88

Grand Contributor
Forum Donor
Joined
Feb 23, 2016
Messages
20,524
Likes
37,057
https://secure.aes.org/forum/pubs/conventions/?elib=20690

Be nice to learn more from those with AES access.

The results are roughly summarized in this blog on Stereophile.
https://www.stereophile.com/content/hi-rez-audio-distinguished-blind-testing

Thanks to @Miska for pointing it out to me on another forum.

Using white noise, and gaussian pulses with young listening subjects around 22 years old, positive results in all, but one versions of the test. Those using gaussian pulse gave p<.0001 results. All, but one test had p<.05 results.
 

pkane

Master Contributor
Forum Donor
Joined
Aug 18, 2017
Messages
5,632
Likes
10,205
Location
North-East
https://secure.aes.org/forum/pubs/conventions/?elib=20690

Be nice to learn more from those with AES access.

The results are roughly summarized in this blog on Stereophile.
https://www.stereophile.com/content/hi-rez-audio-distinguished-blind-testing

Thanks to @Miska for pointing it out to me on another forum.

Using white noise, and gaussian pulses with young listening subjects around 22 years old, positive results in all, but one versions of the test. Those using gaussian pulse gave p<.0001 results. All, but one test had p<.05 results.

The details of this study will be interesting to read. The detection is possibly due to interactions of the high frequencies with the audible range through aliasing and IMD. Noise and simple test signals make such distortions easier to detect; it is much harder to do with music recordings.
 

Sgt. Ear Ache

Major Contributor
Joined
Jun 18, 2019
Messages
1,894
Likes
4,151
Location
Winnipeg Canada
Personally, I don't doubt it's possible to hear audible differences between lossy and lossless/high res even with music under certain pretty highly-controlled circumstances. For me, the issue is only the ridiculous, over-the-top claims from the audiophools about "night and day" differences and "mp3 sucks ass!" and all that. I don't believe any of the claims from people saying they can "always tell when a rogue 320kb mp3 somehow manages to sneak into their pristine lossless play stream." They note that "something is wrong" and when they check, sure enough there's a lossy mp3 playing! Even in their car they can tell! Every. Single. Time. lol...yeah sure you can.

Afaic, if I have to jump through hoops in order to set up a situation where I can somehow actually reliably hear any difference at all between a 320kb mp3 and a lossless file, it's a difference I don't worry about.
 

SIY

Grand Contributor
Technical Expert
Joined
Apr 6, 2018
Messages
10,383
Likes
24,749
Location
Alfred, NY
I have not read it yet (it's bedtime here), but here is the paper...

Thanks, Paul. Two things struck me on my first skim of the paper:

1. "In the ABX test, the participants are not allowed to listen to the stimuli multiple times for each answer. " That seems curious- I've never done or participated in an ABX or triangle test with that restriction.

2. They didn't seem to control for in-band distortion from their electronics and transducers.
 

pkane

Master Contributor
Forum Donor
Joined
Aug 18, 2017
Messages
5,632
Likes
10,205
Location
North-East
2. They didn't seem to control for in-band distortion from their electronics and transducers.

That's what I wanted to see from the paper. I'll get to it tonight or tomorrow, but this would be a really large tell when dealing with ultrasonics, especially at a high magnitude as would be in a white-noise signal. @PaulD , thank you for posting it!
 
OP
Blumlein 88

Blumlein 88

Grand Contributor
Forum Donor
Joined
Feb 23, 2016
Messages
20,524
Likes
37,057
Am I understanding this correctly in that they didn't dither when going from 64 bit float to 16 bit signals? And that the amplitude of the test signals varies up to 1 db between the various sample rates?
 

mansr

Major Contributor
Joined
Oct 5, 2018
Messages
4,685
Likes
10,700
Location
Hampshire
There are several problems here:
  • The paper is not nearly detailed enough to allow independent replication of the experiment. In fact, I only get a vague of idea of what they actually did.
  • They appear to have operated the DAC at different sample rates. Bad idea. There's no telling what spurious differences that might introduce. Better would be to run the DAC at the highest rate and apply a high-quality software low-pass filter with various cut-offs to the test signals.
  • They spend an entire page discussing quantisation to various bit depths but do not mention this further. It is unclear what bit depth(s) the listening tests were performed with. There is no mention of dither, though it probably doesn't matter here, assuming 24-bit resolution was in fact used.
  • "Through these processes, three impulse signals and three white noise signals are made to have magnitude differences within 1 dB." If this means the signals were not level matched (in the (traditionally) audible range), the entire thing goes out the window and down the nearest street drain.
  • The DAC used is a rather mundane model with unimpressive specs. Although not necessarily a huge problem, they really should have recorded the output (both from the DAC and with a microphone) to check for distortion products appearing in the audible range.
  • The spectacular outcome strongly suggests that something is amiss. The paper includes no discussion of possible sources of errors.
  • I could go on...
 

ajawamnet

Active Member
Joined
Aug 9, 2019
Messages
288
Likes
460
From what I'm seeing the total number of participants was 7... hmmm.
part.jpg

That seems like a pretty low sample group...

I've noticed some weird crap that the services do to some of the stuff I've recorded and have done null tests that seem to indicate I'm not making this up. But to really think that most listeners in typical environs even care is a bit of a stretch.... with the trunk lid of their car whumpa whumpa buzz buzz..
 

Juhazi

Major Contributor
Joined
Sep 15, 2018
Messages
1,717
Likes
2,897
Location
Finland
I agree, reports like this shouldn't be published in journals at all, this is not even poster quality! I had AES access for some time and was astonished that actually most of papers are of same inadequate scienticic/statistical quality. Number of participants is so low that no significance can be expected.
 

Soniclife

Major Contributor
Forum Donor
Joined
Apr 13, 2017
Messages
4,500
Likes
5,417
Location
UK
The details of this study will be interesting to read. The detection is possibly due to interactions of the high frequencies with the audible range through aliasing and IMD. Noise and simple test signals make such distortions easier to detect; it is much harder to do with music recordings.
Using a speaker with a single full range driver, and getting more positive results from the speaker than headphones sounds like you might be onto something.
 

Ron Texas

Master Contributor
Joined
Jun 10, 2018
Messages
6,078
Likes
8,914
That they used test signals immediately made me suspicious. All this proves is it is possible to hear a difference under circumstances unlike normal listening and with a particular kind of speaker.
 

PaulD

Senior Member
Joined
Jul 22, 2018
Messages
453
Likes
1,340
Location
Other
I agree, reports like this shouldn't be published in journals at all, this is not even poster quality! I had AES access for some time and was astonished that actually most of papers are of same inadequate scienticic/statistical quality. Number of participants is so low that no significance can be expected.
Yes, the AES has the lowest editorial standard of any journal that I know of, but it's all the audio world has sadly so I keep my sub...
 

ElNino

Addicted to Fun and Learning
Joined
Sep 26, 2019
Messages
557
Likes
722
This paper really could have used editing. For example, it's unclear (at least to me) whether the "impulses" used were actually literal impulses in the signal processing sense, or sinc-filtered impulses.

And what's all the discussion about quantization about, when none of the reported experimental results seem with respect to different levels of quantization? (Unless I'm missing something.)

The whole thing seems very amateurish and makes you wonder if they realized what they were doing.

In-band IMD is a possible explanation with the experimental setup they were using, but I think the most likely explanation is related to the speakers they used. The TD-M1 is a weird active speaker in that it has a non-oversampling, filterless (NOS) DAC mode, which may even be the default. I mean, if you're playing raw impulses into a NOS DAC, you expect to hear something different!

I don't think the idea of using synthetic signals is a bad idea (though you can't necessarily generalize to music, of course)... for example, the absolute phase synthetic signal that Archimago used to successfully ABX absolute phase inversion is legitimately fascinating. But this paper is not a good example.
 

vitalii427

Senior Member
Forum Donor
Joined
Dec 19, 2017
Messages
386
Likes
531
Location
Kiev, Ukraine
Hmm... it’s rather the level of a student coursework than a serious scientific study :confused:
 
Joined
Dec 4, 2019
Messages
37
Likes
21
Personally, I don't doubt it's possible to hear audible differences between lossy and lossless/high res even with music under certain pretty highly-controlled circumstances. For me, the issue is only the ridiculous, over-the-top claims from the audiophools about "night and day" differences and "mp3 sucks ass!" and all that. I don't believe any of the claims from people saying they can "always tell when a rogue 320kb mp3 somehow manages to sneak into their pristine lossless play stream." They note that "something is wrong" and when they check, sure enough there's a lossy mp3 playing! Even in their car they can tell! Every. Single. Time. lol...yeah sure you can.

Afaic, if I have to jump through hoops in order to set up a situation where I can somehow actually reliably hear any difference at all between a 320kb mp3 and a lossless file, it's a difference I don't worry about.

They oddly stop replying when newer lossy codecs are brought up like Opus at 128kb/s. I never understood the anti MP3 view audiophiles have, A lot stop replying when they can't even tell V3 Lame let alone 320k MP3 or ABX proof is asked.
 
Last edited:

LTig

Master Contributor
Forum Donor
Joined
Feb 27, 2019
Messages
5,760
Likes
9,442
Location
Europe
In-band IMD is a possible explanation with the experimental setup they were using, but I think the most likely explanation is related to the speakers they used. The TD-M1 is a weird active speaker in that it has a non-oversampling, filterless (NOS) DAC mode, which may even be the default. I mean, if you're playing raw impulses into a NOS DAC, you expect to hear something different!
Is this really true - a NOS-DAC without reconstruction filter? Of course there will be a difference in the audible range - all signal contents above the Nyquist frequency is folded back below the Nyquist frequency. How can a scientist working in audio not know this? :facepalm:
 
Top Bottom