• 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!

High Resolution Audio: Does It Matter?

D

Deleted member 16543

Guest
You'd probably see some leakage, depending on the filter. 24kHz really has no part in this discussion, except as fs/4, it doesn't really matter except as an anchor point to locate the center of the spectrum.

More appropriate would be to use an actual high frequency cochlear filter, which I have done, but which I can't reveal thanks to the fact it's quite proprietary information.

The results, however, would be slightly worse, in terms of 'a bit more excitation before, and even more after'. That's all I can really say. So I am understating the effect slightly.

Using a high pass (24 to ~48 kHz) type of filter's kernel, processed with sliding Hann windows, would be useful in determining if these possibly audible artifacts that we have just modeled within the audio band, above 16 kHz, are actually audible or not. It's a way to double check the the validity of the model, and to investigate if the use of a filter's kernel (impulse response), rather than its output, is appropriate or not.
 

j_j

Major Contributor
Audio Luminary
Technical Expert
Joined
Oct 10, 2017
Messages
2,267
Likes
4,758
Location
My kitchen or my listening room.
Using a high pass (24 to ~48 kHz) type of filter's kernel, processed with sliding Hann windows, would be useful in determining if these possibly audible artifacts that we have just modeled within the audio band, above 16 kHz, are actually audible or not. It's a way to double check the the validity of the model, and to investigate if the use of a filter's kernel (impulse response), rather than its output, is appropriate or not.
Not sure how this would apply.
 
D

Deleted member 16543

Guest
Not sure how this would apply.
If I try to summarize our discussion so far, we are basically examining Hann sliding windows of various filters' kernels and looking at potential artifacts that may appear below 20 kHz.
So far we have been able to show (potentially) audible artifacts above 16 kHz, based on the modeling of the ear we used (the Hann sliding windows), by examining a sharp brickwall filter at 20 kHz, that has a sinc (or close enough) type of kernel.

Let's now take a signal all comprised between 24 kHz and 48 kHz. No matter what its specific spectrum or the way we created it (naturally recorded ultrasonic sound, sped up recording, high passed recording, etc..), this signal isn't audible.

If it turns out that the sharp transition high pass 24-48 kHz filter has a kernel that produces artifacts below 20 kHz, we have a strong indication that the equivalency between output and kernel can't be used to demonstrate audibility of these artifacts using the sliding windows model. If it did, we could hear "something" when playing a 24-48 kHz signal.
 

j_j

Major Contributor
Audio Luminary
Technical Expert
Joined
Oct 10, 2017
Messages
2,267
Likes
4,758
Location
My kitchen or my listening room.
If I try to summarize our discussion so far, we are basically examining Hann sliding windows of various filters' kernels and looking at potential artifacts that may appear below 20 kHz.
So far we have been able to show (potentially) audible artifacts above 16 kHz, based on the modeling of the ear we used (the Hann sliding windows), by examining a sharp brickwall filter at 20 kHz, that has a sinc (or close enough) type of kernel.

Let's now take a signal all comprised between 24 kHz and 48 kHz. No matter what its specific spectrum or the way we created it (naturally recorded ultrasonic sound, sped up recording, high passed recording, etc..), this signal isn't audible.

If it turns out that the sharp transition high pass 24-48 kHz filter has a kernel that produces artifacts below 20 kHz, we have a strong indication that the equivalency between output and kernel can't be used to demonstrate audibility of these artifacts using the sliding windows model. If it did, we could hear "something" when playing a 24-48 kHz signal.

What we have already shown is that an impulse sent to a proper filter can possibly create a problem. I see no need for further evidence.
 
D

Deleted member 16543

Guest
What we have already shown is that an impulse sent to a proper filter can possibly create a problem. I see no need for further evidence.

I might be alone in this, but I see value in further investigation. While I am intrigued by these possibly audible artifacts, I also want to know if it's worth to add extra IMD to my audio chain just to fix something that "might" be a problem.
If IMD wasn't there I would agree with your safe bet approach of using a gentle filter.

So maybe a possible test could be this. If we play not just any 24-48 signal, but the impulse response of such a filter, can we hear anything?
If we don't, and the plot shows artifacts below 20 kHz, then I would say it's not out of the realm of possibilities that we have a problem in the way we are modeling this phenomenon.
 

j_j

Major Contributor
Audio Luminary
Technical Expert
Joined
Oct 10, 2017
Messages
2,267
Likes
4,758
Location
My kitchen or my listening room.
I might be alone in this, but I see value in further investigation. While I am intrigued by these possibly audible artifacts, I also want to know if it's worth to add extra IMD to my audio chain just to fix something that "might" be a problem.
If IMD wasn't there I would agree with your safe bet approach of using a gentle filter.

So maybe a possible test could be this. If we play not just any 24-48 signal, but the impulse response of such a filter, can we hear anything?
If we don't, and the plot shows artifacts below 20 kHz, then I would say it's not out of the realm of possibilities that we have a problem in the way we are modeling this phenomenon.
We've already inserted a signal that is flat between DC and 48K. That's the most realistic test case.
 
D

Deleted member 16543

Guest
We've already inserted a signal that is flat between DC and 48K. That's the most realistic test case.

I may have missed something, but it seem to me we didn't process the kernel of filters that are flat from DC to 48 kHz (at least not in this discussion).
We processed the kernels of filters that are flat between DC and 20 kHz, then one drops abruptly to -120 dB, the other takes a more gentle slope and is down to rejection around 32 kHz.

Besides, a filter that's flat from DC to 48 kHz may be indicative of the most realistic case, but what I'm trying to do is rather to isolate the potentially audible artifacts due to sharp filtering of ultrasonic frequencies, without having any other signal in the audible band to mask them.

Come to think of it, 24 to 48 kHz came from another line of reasoning I was following:
Since we can't hear any signal whose spectrum has content only above 24 kHz, if we see artifacts below 20 kHz in its plot we know that there might be a problem in the kernel-output equivalency.

Now that it is more clear we are not talking about just any signal, but impulse responses, I think a more fair test would be to process the kernel of a high pass filter that has the first sharp transition band right at 20 kHz, instead of 24.
If we can't hear anything when playing that impulse response, then it stands to reason that those artifacts that the plot shows are even less audible when there is a signal in the audible band masking them.
At least that is my thinking.
 
Last edited by a moderator:

j_j

Major Contributor
Audio Luminary
Technical Expert
Joined
Oct 10, 2017
Messages
2,267
Likes
4,758
Location
My kitchen or my listening room.
I may have missed something, but it seem to me we didn't process the kernel of filters that are flat from DC to 48 kHz (at least not in this discussion).
We processed the kernels of filters that are flat between DC and 20 kHz, then one drops abruptly to -120 dB, the other takes a more gentle slope and is down to rejection around 32 kHz.
A filter that is flat over the entire bandwidth is an impulse.
 
D

Deleted member 16543

Guest
A filter that is flat over the entire bandwidth is an impulse.

I don't see why this would matter.
If you think there are audible artifacts when rejecting all frequencies above ~20 kHz with a sharp filter, the same should be true when rejecting all frequencies below that and keeping the ultrasonic content.
Except that by removing the audible content there's nothing to mask the artifacts, this time. So it should be evident if the artifacts are actually there or not.
 

j_j

Major Contributor
Audio Luminary
Technical Expert
Joined
Oct 10, 2017
Messages
2,267
Likes
4,758
Location
My kitchen or my listening room.
I don't see why this would matter.
If you think there are audible artifacts when rejecting all frequencies above ~20 kHz with a sharp filter, the same should be true when rejecting all frequencies below that and keeping the ultrasonic content.
Except that by removing the audible content there's nothing to mask the artifacts, this time. So it should be evident if the artifacts are actually there or not.

Perhaps as a subjective test, but not as anything else.
 
D

Deleted member 16543

Guest
Perhaps as a subjective test, but not as anything else.

A subjective test might be the best we can do.
How else would you investigate the audibility of these artifacts if not by having people listen to a signal that isolates them, and reporting if they can hear anything or not?
 

j_j

Major Contributor
Audio Luminary
Technical Expert
Joined
Oct 10, 2017
Messages
2,267
Likes
4,758
Location
My kitchen or my listening room.
A subjective test might be the best we can do.
How else would you investigate the audibility of these artifacts if not by having people listen to a signal that isolates them, and reporting if they can hear anything or not?

Well, in order for this to be meaningful, we'd need a few hundred subjects, pre-tested and screened, will callbacks for people who show sensitivity to prove statistical significance.

I wish I had the budget.
 
D

Deleted member 16543

Guest
Well, in order for this to be meaningful, we'd need a few hundred subjects, pre-tested and screened, will callbacks for people who show sensitivity to prove statistical significance.

I wish I had the budget.

Yes, I hear you.
Also, the test should be performed with some kind of super tweeter with no resonance above 20 kHz.
Speakers are pretty much always the bottleneck when trying to investigate miniscule artifacts audibility.

But for now the people reading this thread and their speakers would be a good enough start, I think.
 

Pluto

Addicted to Fun and Learning
Forum Donor
Joined
Sep 2, 2018
Messages
990
Likes
1,631
Location
Harrow, UK
What interested me, and what I was surprised to find, was how edits/splicing appears to have been done at 7ips, and not 15
Having looked at the video in question, the ¼" tape (on the Studer C37) was running at 15ips, not 7½ips. The latter would have been considered acceptable only for office listening copies.
 

j_j

Major Contributor
Audio Luminary
Technical Expert
Joined
Oct 10, 2017
Messages
2,267
Likes
4,758
Location
My kitchen or my listening room.
Yes, I hear you.
Also, the test should be performed with some kind of super tweeter with no resonance above 20 kHz.
Speakers are pretty much always the bottleneck when trying to investigate miniscule artifacts audibility.

But for now the people reading this thread and their speakers would be a good enough start, I think.

For this test, I know of some tweeters that would be fine, with a bit of FIR correction in front of them, they can be gotten to well under .2 dB ripple from 5k to 30K. That's the easy part here. (since we already have those running in all of our demo rooms)
 

djgras

Member
Joined
Jun 9, 2022
Messages
37
Likes
8
High Resolution Audio: Does It Matter?
By Amir Majidimehr

[Note: This article was published in the January 2015 issue of Widescreen Review Magazine. This is a slightly revised edition.]

Yes!

Thought I give you the short answer first in case you are impatient like me.

If you have read my past articles you know that I have a passion for excellence in engineering when it comes to audio. Invariably the question is raised as to why. Does that precision matter? How about High-resolution Audio? I hope to answer this question in this article or as a minimum provide an objective path for you to decide for yourself.

Background
Go into any audio forum and you immediately see food fights over the merits of distributing audio at higher resolution than the CD. The two warring camps are quite polarized. One goes as far as the CD spec is “more than good enough” with 16 bit per sample resolution and sampling rate of 44.1 KHz. You will often see proclamations such as music not even needing more than 12 to 13 bits. And that most listeners can’t hear 20 KHz so the 22.05 KHz bandwidth represented by CD’s sampling rate is more than sufficient.

The other crowd is the “night and day” camp. They say that if you can’t hear the difference you must not have a good enough system, ears or both. And that the benefits of high resolution audio scale more or less with its spec. If 96 KHz is good, then 192 KHz must be even better.

Each camp is content with its position in this matter. The good enough camp points to what it says is “scientific evidence” in the form of a listening test from Meyer and Moran. As I explain later that report states that improvements from higher sampling rate/bit depths than the CD are not audible.

Who is right here? Is the difference as big as the numerical difference between specs of the CD and high resolution audio? Or is it that there are no audible differences regardless of technical and measured differences? In my opinion both camps are wrong! Extreme positions in any topic become hard to defend and nowhere is it more true than in this argument.

The fallacy of the good enough argument is that it relies on a limited number of tests performed almost entirely by hobbyists who don’t have the necessary skills, experience, knowledge and tools to perform these tests properly. The night and day camp likewise relies purely on sighted tests which unfortunately are subject to severe placebo effect as one is easily convinced to hear better sound if its specs are better.

The reason these discussions resemble battles is because neither side tries to convince the other using their own evidence. Good enough camp demands double blind test data which the other camp dismisses as being an improper method of evaluation. Likewise the night and day camp can’t possibly convince the other that the only path is “to just listen” with eyes wide open.

So how do we make forward progress? My approach is to focus on the good enough camp. They are the ones that take the position of listening to “scientific evidence” and believe in the validity of double blind tests and the science of audio. What is about to follow is such a presentation. I will present data and research that should as a minimum give pause to people in the good enough camp. That their extreme position is not defensible.

Please allow me to also state one of my pet peeves in this discussion. There is nothing magical about 16/44.1 specification of the CD. It is not like extensive listening tests and research were performed to pick these values. They were selected by Sony and Philips in the creation of CD to balance the recording capacity and fidelity. So at some level the one camp is defending an arbitrary set of numbers.

High Resolution Audio
Let’s step back and define what we mean by high resolution audio. There is no formal definition so I will resort to my own: anything above CD’s 16-bit/44.1 KHz in my book is high resolution audio. The most common step above that which is used frequently in video production is 16 or 24-bit samples at 48 KHz.

Fans of high resolution audio no doubt want to see much bigger numbers than 48 KHz. But 48 KHz with the original samples of 16 or 24 can still be beneficial.

Mind Your Business
In this article you will see me not only talking about technical aspects of this discussion but also the business component. It is the latter that ultimately determines why we should get behind high-resolution audio and more so than any technical points.

For now let’s address one of the most important considerations in this topic:the CD format is on its way out. It is just that it doesn’t know it yet! No, don’t go citing music industry statistics of billions of CDs still being sold. I know that but it doesn’t matter. CD will in the ensuing years find itself without a customer.

The music market is all about the mass consumer. Enthusiasts like us don’t even account for round off error. As everyone knows, the mass market has strongly adopted online consumption of music. Compressed MP3/AAC audio can easily be streamed or downloaded giving instant access to millions of music tracks. Yes, there is a fidelity drop but to the mass market music buyer that difference is either inaudible, not important or both.

To many younger buyers the notion of buying a physical CD is a foreign and antiquated concept. They can’t play the CD in their smartphone. And who wants to wait for the CD to arrive, rip it and then play it on the run? They certainly don’t. And unlike Vinyl there is nothing nostalgic about the CD. Not yet anyway. As this group ages this becomes the norm and the CD will find itself without a mass market consumer.

What this means is that the people who fund the compensation of music labels and royalties of the talent have already voted with both feet that the CD is destined to go away as a format. To be sure CD consumption is not stopping anytime soon or ever. But the trend is clear and unstoppable.

There is evidence of this already. I love modern music but increasingly find that such music has gone directly to MP3/AAC on iTunes/Amazon, bypassing the CD format. The trend is accelerating in that I run into at least one album a week that I can no longer find in CD format. Yes, this is specific to my listening habits but it is indicative of the future.

If you are a new artist you are not going to stamp CDs when you can just submit your music to the Amazon and iTunes of the world and be done with it. Who wants to deal with inventory of physical goods in an uncertain market?

Fortunately a new distribution channel has been opened for enthusiasts in the form of high resolution audio. More and more content is being distributed in this manner, bringing with it the same convenience of instant downloading that the mass market enjoys with MP3/AAC compressed formats. The market is tiny but over time I believe it will continue to grow and become the destination format for enthusiasts especially as the CD format starts to shrink as a viable alternative.

Fast forward three, five or more years and you can easily see how the CD falls into no man’s land. Masses don’t want it because they would have completely shifted to online compressed formats. And audiophiles would have moved on to high resolution audio for its better spec if not fidelity. CD will find itself without a strong need.

Currently there is one important barrier to this future: cost. Comparison of most albums in high resolution format from the specialty online retailers and mass market CD outlets like Amazon shows a significant premium. The reason for this is one of the untold secrets of the music business. Namely the fact that CDs are often sold at a loss in order to get the customer to come into the store to buy something else that has a profit margin. Since new CDs (and DVDs/Blu-rays) are constantly released, they act as a strong factor in getting a customer to visit your store often. People don’t buy microwave ovens or cameras every week or two but do consume music that way.

Single-category high resolution audio distributors don’t and can’t have the above business model as they have nothing else to sell. Their only profit margin is that of the music content itself so they have to charge proper profit margin to stay in business, hence the higher prices. Fortunately the high-end of the music market is not as price sensitive so this is not a big barrier to the future I have described.

f635d4_82a07e5ad76d43b1a7122c6c53a8dc9d~mv2.png

Figure 1: Meyer and Moran test of audibility of SACD/DVD-A
converted down to 16/44.1 using a CD-ROM recorder.

But Is It Audible?
Breathe a word about high resolution audio and you are immediately challenged to prove whether the difference between it and CD is audible. Before you can open your mouth, you are read the findings of a test (“engineering report” in AES parlance) in 2007 by Meyer and Moran, published in the Journal of Audio Engineering Society titled,Audibility of a CD-Standard A/D/A Loop Inserted into High-Resolution Audio Playback. The test involved using a DVD Audio or SACD player as the source (remember this is at the height of that format war) with the audio split into two paths. One went straight into an ABX switch box and the other, through a CD Recorder’s monitoring loop. That loop converted the analog output of the DVD-A/SACD players into digital, presumably at 16/44.1 KHz and then back to analog on the way out (see Figure 1).

User then used the ABX switch box to play A (e.g. the DVD-A/SACD) without modification, followed by B (the loop that forced the conversion down to 16/44.1 through the CD recorder) and then had to identify X to be more like A or B.

Since in any such “forced choice” test the listener can simply guess, a protocol is used to differentiate between that and actual fidelity difference. The statistical method calls for determining the probability of chance being lower than 5% that the person was guessing randomly. Or put inversely that there is 95% confidence in the person hearing an audible difference between A and B.

The headline from the report was that out of all the trials, the number of correct guesses was only 49.82% (why anyone would report such numbers with two decimal places where the margin of error is quite a bit bigger than this is beyond me). The good enough camp happily runs with this summary by declaring that the difference between DVD-A/SACD and its “CD” version was no better than chance, ergo there is no audible difference.

Lost in that is one tester who managed to get 8 out of 10 right meaning there was 94.5% probability that he was identifying the proper source and not guessing. This is so close to 95% threshold that it should have been noted as significant and countering the larger conclusion but was not. Two other testers managed 7 out of 10 correct selections. These were all dismissed as exceptions and the total number of trials/listeners incorrectly relied upon.

Beyond the above, there are some serious errors in the Meyer and Moran test:

1. Meyer and Moran failed to test their source material to see if they indeed had spectrum that would be captured by higher sampling rates of SACD/DVD-A. Some of the titles they used were actually upsampled CD masters so they were one in the same when converted down to 16/44.1. This is a very basic mistake. Every test of this sort must include verification of the assumptions. Namely that the content being used is what it is thought to be. That they did not see fit to do so casts a very dark shadow on the quality of this test.

2. They do not provide any measurements that demonstrate that the CD-ROM recorder used as a poor man’s sample and bit depth converter to 16/44.1, actually performed that assumed function. The Journal report does not even mention what that equipment was. Later online posting revealed that it was an HBB CD-ROM recorder as I have noted in Figure 1. The manual for that unit fails to say anything at all about the monitoring loop that was used in the test for down conversion. It is possible that it is an analog pass-through path and hence no conversion to 16/44.1 occurred. The authors insist that they had measurements that proved otherwise but none are provided for review in the paper or online. This is no way to perform a “scientific study.” All important assumptions need to be verified and documented. Next to content itself being high resolution, nothing is more important than showing the response of this down conversion being that of 16/44.1 KHz.

3. There were no controls. Industry and research best practices and international standards such as ITU-R BS1116 require that there be controls that are used to verify proper operation of the test harness. What is a control? It is a stimulus with a known outcome. An example in this scenario would be down sampling all the way down to 22 KHz with the bandwidth extending to only 11 KHz. If the listeners still report random results, then we know something is wrong. Maybe the connections to the ABX switchbox are mis-wired. Or data gathering. Humans make mistakes and we need ways to catch those mistakes.

4. Another important use of controls is screening out testers who do not possess critical listening abilities for the test in question. If our test fixture is correct but listener still votes randomly, we must eliminate them from the listening panel. Having these testers participate still would serve to dilute the results. And is a sure way to get "no better than chance" outcome since that is precisely what these non-critical listeners do.

The above is an example of a statistical concept called the Simpson’s Paradox. It says that improper summation of a group of results can create false conclusions Here, it matters not that one hundred people could not hear the difference,if five could. If those five reliably found a difference then we know the difference is there. Mixing the other one hundred in there will only serve to generate incorrect data, not strengthen it.

5. Meyer and Moran tests was performed using multiple sets of playback hardware, content and source equipment. These make for different tests whose outcomes cannot be combined into one statistic. This violates basic principles of statistics. This failing was raised in a letter to the Journal of AES by Dr. Dranove. Meyer and Moran respond by accepting the criticism in their response:


Dr. Dranove has set requirements for our engineering report that were not part of our plan, and then dismissed it for failing to meet them. In hindsight it probably would have been better for us not to cite the total number of trials as there are issues with their statistical independence, as well as other problems with the data. We did not set out to do a rigorous statistical study, nor did we claim to have done so. Accordingly it may not mean much to do a more detailed data analysis, though we have done further work on it that we will discuss later.

And:

We did not know in advance what source material, what type of system, or which subjects would be the most likely to reveal an audible difference.

Ad-hoc testing combined with less than proper statistical foundation does not make for authoritative work.

If we are going to rely on conclusions of such tests, then we better apply proper rigor to them and follow best practices. Take testing of audio compression. Critical music segments are used that are revealing of compression artifacts, not some random set of music files or even “audiophile tracks.” We utilize expert listeners (incorrectly called “Golden Ears”) who are trained on how audio compression works and the type of artifacts it introduces. While testing is sometimes performed using a larger audience that is not trained, critical design decisions and competitive studies always follow these practices. Controls are routinely used to screen out poor listeners and protocol errors. Knowledge of what is being tested determines how the test is created. All the things that were not done in Meyer and Moran test.

Yes, my criticism is harsh and strong. Just because we have a pair of ears and access to a switchbox doesn’t mean we are qualified to run off and perform such tests. This is a specialized field that requires experience and expertise to perform tests correctly. Clearly that was not what governed this testing.

Do we throw out all of Meyer and Moran results then? No, that would not be proper either. If there were night and day audible differences it should have showed up better than it did. The fact that it did not means that these differences, if they existed in their setup, were small. So this work is also a cautionary note to the other camp.

But Is It Audible, Part 2
For some seven years the Meyer and Moran test has been the only “published peer reviewed” test of its kind. That changed at the AES conference in 2014 where the results of another listening test contradicted its outcome.

The listening tests were conducted by Helen M. Jackson, Michael D. Capp, and J. Robert Stuart in a paper titled:the audibility of typical digital audio filters in a high-fidelity playback system. Bob Stuart is the founder of Meridian Audio. Same company that produced the MLP lossless audio that became mandatory in DVD Audio and part of the specification in Blu-ray in the form of Dolby TrueHD. Bob is an AES Fellow with a background in signal processing and psychoacoustics. Given that, you can expect this work to be more authoritative than Meyer and Moran and it is.

The authors took a commercial high resolution track, Haydn's String Quartet Op.76 No.5 in D “Finale, Presto" from “Nordic Sound (2L Sampler)" and reduced its resolution six different ways:

1) Filtering down to 22.05 KHz representing 44.1 KHz sampling while keeping the bit depth the same as the original (24 bits).

2) Filtering down to 24 KHz representing 48 KHz filtering (again at 24 bits).

3) Filtering down to 22.05 KHz with bit depth also reduced to 16 bits but with no dither (i.e. just throwing out the extra bits of resolution).

4) Filtering down to 24 KHz and 16 bits with no dither.

5) Filtering down to 22.05 KHz with bit depth reduced to 16 bits with dither.

6) Filtering down to 24 KHz with bit depth reduced to 16 bits with dither.

f635d4_0460e72c22a243559d9c756b7a3db987~mv2.png

Figure 2: Stuart et al. listening test results relative to 95% confidence (dashed line)

In simpler terms they tested what happens when you change the sample rate by itself with and without conversion to 16 bits using different schemes. Eight (8) listeners participated at different ages. None were trained but the test properly included a training phase as recommended in ITU BS 1116.

Total number of 160 trials were conducted across all listeners. The 95% confidence that results are not due to chance require number of correct responses to be above 56% right answers (yes, 56%; see this article onStatistics of ABX Testing.htmlfor why). As Figure 2 shows, all but one test exceeded this target. The one test that didn’t quite make it there (at 48 KHz) did so when the results of the critical segments in the music track were counted.

The results show that mere filtering of the high sample rate track to that of CD's 44.1 KHz and video’s 48 KHz had audible effect (to the stated statistical significance).

Let's note that none of the testers could hear above 20 KHz and some probably could not even get that far. How is it that filtering the ultrasonics that testers could not hear was audible? The answer is that we introduced audible artifacts, not that we removed what we could not hear in the first place.

The paper authors hypothesize that it is the filtering “ringing” in the time domain that may have caused these artifacts. Ringing is an unavoidable manifestation of digital filtering. The sharper the filter has to be, the longer its “tails” (ringing) in time domain. The ear is not a spectrum analyzer but rather "hears" the samples as they arrive. In theory the ringing can be audible and potentially is what was heard.

The Stuart et al. test used a professional tool called Matlab which is routinely used to simulate different signal processing algorithms (think of it as Excel for Accountants). To probe the ringing effect, I used the same tool to create a filter similar to what they used in the test:

f635d4_d3807a7d43ca4a7e96773accbbf99f1d~mv2.png


As you can see from the filter frequency (“Magnitude”) response we have a sharp exclusion of frequencies above 24 KHz (this is for 48 KHz sampling). Anyone looking at this graph would not think of any audible effects given the ruler flat response in the audible band to 20 KHz. Another picture emerges however when we look at the filter response as it operates on our music samples in “time domain:”


f635d4_fb6edee5df184f3fad4650eb5e88a558~mv2.png


As I have noted on the graph, there are oscillations called “ringing” that go on before our sharp transient in the middle and after it. The ringing after the transition is not usually audible as it is masked by the loud sound of the transition. The ones prior to it, called pre-echo, are not so fortunate. If there is sufficient energy there, it may just be audible.

Let’s compare the above filter to one that has far gentler slope:

f635d4_dcc8f23d261140d488d685ce2a665c5e~mv2.png


Notice how the pre-ringing is essentially gone in about 0.7 milliseconds (0.9 msec – 0.2 msec) whereas in the sharp filter case, it went on for more than 3 milliseconds. The need for such sharp filters exists because CD's 44.1 KHz and Video's 48 KHz don't allow a lot of room for gentle roll off. Whereas with high sampling rate we can have very gentle filters that avoid this issue as a practical matter.

Again, this is a theory and not necessarily verified in this test but is one of the best explanations we have as to why there was an audible effect.

What did the subjects really hear? Here is how they described the differences in filtering:

It was reported that filtering gave “softer edges" to the instruments, and “softer leading edges" to musical features with abrupt onsets or changes. Echoes, when audible, were identified as being affected the most clearly by the filtering. It was felt that some of the louder passages of the recording were less aggressive after filtering, and that the inner voices (second violin and viola) had “a nasal quality." Overall, the filtered recording gave a “smaller and flatter auditory image,” and specifically the physical space around the quartet seemed smaller.

The effect of conversion from 24 bits to 16 bits (“quantization”):

Listeners described that quantization gave a “roughness" or “edginess" to the tone of the instruments, and that quantization had a significant impact on decay, particularly after homophonic chords, where “decay was sustained louder for longer and then died suddenly." This could be an effect of quantization distortion; it is interesting that this was audible even in a 24-bit system, and is consistent with the hypotheses of Stuart [29] that 16 bits are not sufficient for inaudible quantization.

More Audible Differences
You may have heard of the practice of loudness compression. This is a type of processing that is applied to music to make its average loudness higher. Its justification in the past was for radio broadcast where the label/talent wanted to get the attention of the listener in the crowded set of commercials and other music tracks. Today’s reasoning for its existence is to make one’s song not sound too soft in a playlist of MP3s.

Whatever the justification, the huge amount of loudness compression applied is an enemy of high fidelity. Perfectly good stereo high resolution mixes are degraded sonically in the process of mastering the bits for CD/MP3/AAC distribution. Dynamics are lost and music no longer sounds enjoyable.

There have been loud cries from enthusiasts and recording engineers alike against loudness compression to sadly little effect. The labels and talent pay for the production of music and what they want, is what gets produced. Since the masses are responsible for almost all of the revenues of music, what they want overrides everyone else's desires.

Distribution of high resolution audio to consumers gives us a path to circumvent this business practice. The audience for high resolution music are enthusiasts who do not want loudness compression. To the extent there is a market for such content, then the label/talent would have no reason to insist on loudness compression for high resolution content. Remove that from the typical CD and we get instant fidelity improvement. Let’s read the Meyer and Moran report on this point:

Though our tests failed to substantiate the claimed advantages of high-resolution encoding for two-channel audio, one trend became obvious very quickly and held up throughout our testing: virtually all of the SACD and DVD-A recordings sounded better than most CDs—sometimes much better. Had we not “degraded” the sound to CD quality and blind-tested for audible differences, we would have been tempted to ascribe this sonic superiority to the recording processes used to make them. Plausible reasons for the remarkable sound quality of these recordings emerged in discussions with some of the engineers currently working on such projects. This portion of the business is a niche market in which the end users are preselected, both for their aural acuity and for their willingness to buy expensive equipment, set it up correctly, and listen carefully in a low-noise environment.

Partly because these recordings have not captured a large portion of the consumer market for music, engineers and producers are being given the freedom to produce recordings that sound as good as they can make them, without having to compress or equalize the signal to suit lesser systems and casual listening conditions. These recordings seem to have been made with great care and manifest affection, by engineers trying to please themselves and their peers. They sound like it, label after label. High-resolution audio discs do not have the overwhelming majority of the program material crammed into the top 20 (or even 10) dB of the available dynamic range, as so many CDs today do.


f635d4_4cf06b2096604ec58f5a8038f7318b8f~mv2.jpg


For a rather current example of this, buy the excellent (but overplayed)Adele: Live at the Royal Albert Hall combo Blu-ray and CD. The CD is of course at 16/44.1. The Blu-ray version has modest spec of 48 KHz by high-resolution standards. Yet the subjective difference is night and day. The Blu-ray recording is nice to listen to whereas the CD quite horrid. Play them both in two different players and switch between them and you immediately hear how much harm is done in the production of the CD master.

No, the situation is not perfect. There is no guaranteed that high-resolution content is not just a remastered CD with loudness compression. Or that it is high-resolution but has loudness compression. Same is true of the content itself. You can get bad music or good music. Seek our online reviews before trusting high-resolution tracks to be of higher fidelity.

Summary and Final Thoughts
As I mentioned at the outset, high resolution audio makes a difference. And a huge one at that in the way it gives us access to stereo masters prior to re-mastering for the CD. That path means the music can be free of loudness compression which will have clear benefit, putting aside any additional sonic fidelity due to use of higher bit depths and sampling. Given the fact that CD has no choice but to go away in the future, we as enthusiasts better get behind high resolution audio distribution. Nothing but goodness comes from having more choices of formats for our music.

References
"Audibility of a CD-Standard A/D/A Loop Inserted into High-Resolution Audio Playback,"E. Brad Meyer and David R. Moran, Engineering Report, J. Audio Eng. Soc., Vol. 55, No. 9, 2007 September

Comments on“Audibility of a Cd-Standard A/D/A Loop Inserted Into High-Resolution Audio Playback,”LETTERS TO THE EDITOR, J. Audio Eng. Soc., Vol. 58, No. 3, 2010 March

“The audibility of typical digital audio Filters in a high-fidelity playback system,”Convention Paper, Presented at the 137th AES Convention 2014

“Methods for the subjective assessment of small impairments in audio systems,”ITU Recommendation BS.1116-2

"Statistics of ABX Testing.html,"Amir Majidimehr
Hi Amir, thanks for all the reviews and work.
I am hoping for a 192 kHz miniDSP SHD STUDIO 2 to go USB between my fanless win11 PC (Audirvana Studio with local files on SSD, Qobuz) to the Aqua Hifi La Scala dac in a 2.1 or 2.2 system (Gryphon amp, Raidho bookshelf speakers).

There is also the question of what miniDSP does to the original sampling rate?
Best regards, Didier
 

sofrep811

Active Member
Joined
Jun 4, 2016
Messages
253
Likes
319
Missed this several years ago when posted. Excellent work, Amir!! I own an Oppo and still buy SACD's and Blu-Ray titles.
 

dtaylo1066

Addicted to Fun and Learning
Joined
Jan 12, 2019
Messages
628
Likes
741
The limited number of Hi-Res files I have to listen to sound quite good on my modest system: Hypex Amps, Emotiva DACs, TinkerBoard with Volumio, DIY SEAS monitors, my CDs stored on NAS and streamed over Volumio.

Part of that is that I suspect that particular care was taken in the recording, mixing, etc. of the Hi-Res music, many times created by lesser known artists. I have some Redbook CD's that sound fantastic, and others sound like excrement. A lot has to do with how the record was produced.

When Sony and Phillips determined the specs for CD's it was a very different world. The CD fairly rapidly diminished vinyl. Now streaming has diminished the CD, and vinyl has made a comeback, as has cassette, IMHO more out of nostalgia than anything else.

The reason Hi-Res has not taken off is that the masses who consume music do not listen critically. Lossy formats are good enough for them to blast in their earbuds or in their cars.

It is only a small percentage of us audio freaks who closely and critically listen to music. Now that Hi-Res and streaming have become affordable enough for a slow growth in distribution to occur, I welcome it.

If one guy picked it right 8 out of 10 times, that is meaningful. Given the average person's hearing, the average person may never be able to detect a delta between lossy, Redbook or Hi-Res. If some can, and I think I am one of them, then it is worth it.
 

Blumlein 88

Grand Contributor
Forum Donor
Joined
Feb 23, 2016
Messages
20,521
Likes
37,050
The limited number of Hi-Res files I have to listen to sound quite good on my modest system: Hypex Amps, Emotiva DACs, TinkerBoard with Volumio, DIY SEAS monitors, my CDs stored on NAS and streamed over Volumio.

Part of that is that I suspect that particular care was taken in the recording, mixing, etc. of the Hi-Res music, many times created by lesser known artists. I have some Redbook CD's that sound fantastic, and others sound like excrement. A lot has to do with how the record was produced.

When Sony and Phillips determined the specs for CD's it was a very different world. The CD fairly rapidly diminished vinyl. Now streaming has diminished the CD, and vinyl has made a comeback, as has cassette, IMHO more out of nostalgia than anything else.

The reason Hi-Res has not taken off is that the masses who consume music do not listen critically. Lossy formats are good enough for them to blast in their earbuds or in their cars.

It is only a small percentage of us audio freaks who closely and critically listen to music. Now that Hi-Res and streaming have become affordable enough for a slow growth in distribution to occur, I welcome it.

If one guy picked it right 8 out of 10 times, that is meaningful. Given the average person's hearing, the average person may never be able to detect a delta between lossy, Redbook or Hi-Res. If some can, and I think I am one of them, then it is worth it.
Why do you think that you are one of them?
 
Top Bottom