Skip to main content

Notice

Please note that most of the software linked on this forum is likely to be safe to use. If you are unsure, feel free to ask in the relevant topics, or send a private message to an administrator or moderator. To help curb the problems of false positives, or in the event that you do find actual malware, you can contribute through the article linked here.
Topic: The Audibility of Typical Digital Audio Filters in a High-Fidelity Pla (Read 6291 times) previous topic - next topic
0 Members and 1 Guest are viewing this topic.

The Audibility of Typical Digital Audio Filters in a High-Fidelity Pla

In November HA hosted  a noisy if not informative discussion of a recent AES Conference paper:

"The Audibility of Typical Digital Audio Filters in a High-Fidelity Playback System"

I think that most of the sources of noise on that thread are busying themselves elsewhere, so I hope that the following news items will not stir things up too much.

The AES invites online commentary on its publications and meetings. For the paper in question, it is anchored here:

AES Comments log for this paper

What I think is a well-thought out and complete comment can be found here:

Stefan Heinzmann's 12/19/2014 comment

"
I see a number of problems with the paper, some of which are:

The conclusions in the abstract and in the introduction differ markedly from those offered at the end of the paper. They are also not adequately supported by the research presented in the paper.
The introduction contains a lengthy series of speculations which seem to be preoccupied with casting doubt on some past research, yet the paper essentially fails to substantiate the speculations. This comes dangerously close to being unfair.\

The criticism of the ABX test procedure that is offered in the introduction is poorly justified. The "cognitive load", as called by the authors, is entirely under the control of the listener in an ABX test, since the listener selects when to switch and what to switch to. There is no requirement to keep all three sounds in memory simultaneously, as criticised by the authors. Consequently, it is unclear what advantage the method chosen by the authors offers over an ABX test. Furthermore, the informal use of the term "cognitive load" seems to suggest tacitly, that a higher "load" is detrimental to the ability to distinguish between different sounds. I'm not aware of any study that confirms that. Indeed, one could just as easily suspect the opposite, namely that the availability of more sounds would increase this ability. Neither of those suggestions can of course be taken for granted. The authors shouldn't appeal to their interpretation of common sense when criticising a test method, and rely on testable evidence instead.

The quantization to 16 bit was accompanied by either no dither, or RPDF dither. As the authors rightly state, neither is satisfactory, a well-known fact. However, using nonoptimal test conditions defeats the aim of showing deficiencies of the CD-format itself. If artefacts are uncovered with this setup, they may be attributed to the test conditions rather than the format. This means that the conclusion drawn by the authors, regarding the CD-format, is unjustified by the result of their research. If the CD-format as such is under scrutiny, the aim must be to remove all other factors as much as possible. This rings especially dissonant with the authors' criticism of the work of others offered in the introduction.

While the usage of no dither or RPDF dither was justified with a reference to alleged deficiencies of some real-world implementations of converters, the authors chose unusually high cut-off frequencies in their lowpass filters. This can only make it more likely that any uncovered artefacts are due to the test conditions, and not due to the CD format. Again, the authors ought to make clear whether they want to address deficiencies of real world implementations, and if so direct their conclusions at those implementations, or whether they want to address deficiencies of a format, in which case they should attempt to exclude or reduce the deficiencies of real-world implementations from their test setup as much as possible.

Hence I conclude: The research that is presented in this paper shows evidence that supports conclusions 1 and 2 at the end of the paper. I don't see how it supports conclusions 3 and 4, which appear speculative to me. Conclusion 5 isn't actually a conclusion, it rather seems to describe a preconception of the authors, which affected their design of the test procedure. I don't see any attempt on their part to investigate to what extent this preconception is actually valid.

Of the two main conclusions offered at the end of the abstract and the end of the introduction, neither are supported by the research, indeed the tests conducted were not designed to address those questions. That doesn't make the conclusions automatically false, but it casts serious doubts on the authors' interpretation of their own findings.
"

Comments?

The Audibility of Typical Digital Audio Filters in a High-Fidelity Pla

Reply #1
Why does AES does not have a functional peer-review before publication?

Basically all the critiques brought forth in the comment have been stated on HA already, and anybody honest and half-knowledgabe in the audio field should have been able to spot the most obvious deficiencies.
It's only audiophile if it's inconvenient.

The Audibility of Typical Digital Audio Filters in a High-Fidelity Pla

Reply #2
Why does AES does not have a functional peer-review before publication?


I think that is a very good question. The accolades heaped on this paper before its public presentation seem to have have created a serious credibility problem for some.

Quote
Basically all the critiques brought forth in the comment have been stated on HA already, and anybody honest and half-knowledgeable in the audio field should have been able to spot the most obvious deficiencies.


Agreed.  I think that Mr. Heinzmann did a nice job of forming the ideas that many have developed on their own, especially with American English probably not being his first language.

CV info for Mr. Heinzmann

The Audibility of Typical Digital Audio Filters in a High-Fidelity Pla

Reply #3
I even go so far that this paper did cost a lot of the AES reputation. It looks like a Buy-In by Meridian not only to me.
Is troll-adiposity coming from feederism?
With 24bit music you can listen to silence much louder!

The Audibility of Typical Digital Audio Filters in a High-Fidelity Pla

Reply #4
I even go so far that this paper did cost a lot of the AES reputation. It looks like a Buy-In by Meridian not only to me.


I think that this document may have something to do with it:

Dolby TrueHD White paper

The document does not cut and paste for me, so interested parties will have to link it for themselves. I think that the operative words are Dolby, Meridian, and Apodizing.

For a conventional technical view of apodizing:

Apodizing Expose Link

The Audibility of Typical Digital Audio Filters in a High-Fidelity Pla

Reply #5
Quote
The criticism of the ABX test procedure that is offered in the introduction is poorly justified.

Arny,

I'll bet when you invented ABX you never expected it to be so "controversial"!     

Do you ever feel like Copernicus or Galileo? 

Doug.

The Audibility of Typical Digital Audio Filters in a High-Fidelity Pla

Reply #6
Quote
The criticism of the ABX test procedure that is offered in the introduction is poorly justified.

Arny,

I'll bet when you invented ABX you never expected it to be so "controversial"! 


The background on ABX is that it was developed in a turbulent sea of controversy to help quell that controversy.

The controversy at hand in the mid 1970's was over whether or not Quasi Complementary power amps necessarily sounded worse than the newer Full-Complementary ones. Henry Pearson said yes, and his willing disciple Arny was very willing to straighten out the wretched non-believers in The Absolute Sound among his local hi-fi club members.

Quote
Do you ever feel like Copernicus or Galileo?


No. I didn't start this fire. ;-)

The Audibility of Typical Digital Audio Filters in a High-Fidelity Pla

Reply #7
What I think is a well-thought out and complete comment can be found here:

Stefan Heinzmann's 12/19/2014 comment
Something was detected blind. Possibly the filtering. I think this is about as far as we got:
http://www.hydrogenaud.io/forums/index.php...st&p=881245

Also, for those who have forgotten...
http://www.hydrogenaud.io/forums/index.php...st&p=883326
http://www.hydrogenaud.io/forums/index.php...st&p=883151

Cheers,
David.

The Audibility of Typical Digital Audio Filters in a High-Fidelity Pla

Reply #8
What I think is a well-thought out and complete comment can be found here:

Stefan Heinzmann's 12/19/2014 comment
Something was detected blind. Possibly the filtering. I think this is about as far as we got:
http://www.hydrogenaud.io/forums/index.php...st&p=881245'


Trouble is that if there was something that was heard, it was contrived by doing a lot of things suboptimally or just plain wrong.

In essence, the  paper is composed of several straw man arguments:

The first straw man argument was that the ABX test has inherent but avoidable problems with sensitivity. This was based on a criticism of a test called ABX that was first published in a 1950 JASA article.  It turns out that the test that the JAES called ABX in 1982 is a different test that was developed in full knowledge of the 1950 paper, but was designed to overcome a number of readily observable difficulties when applied to tests of audio gear for music reproduction.  Between the 1950 ABX test and the 1982 ABX test there were a number of generations of modifications to the 1950 test, with generally improving applicability to the situation at hand.  The test that was set forth in the 1982 JAES article by Clark was the final iteration of that process, and had been in use by the author and his associates for about 5 years. It stands to this day. Why the authors of the paper we're discussing became confused about which ABX test they were criticizing is hard for me to imagine.

The second straw man argument relates to the simulation of "Typical Digital Audio Filters". One might think that one would base any criticism of "Typical Digital Audio Filters" on real world typical digital audio filters, but instead a composite of numerous well-known bad digital audio practices that beginners are taught to avoid  was settled on. Why the authors of the paper we're discussing became confused about which digital filters  they were criticizing is again hard for me to imagine.


Quote
Also, for those who have forgotten...
http://www.hydrogenaud.io/forums/index.php...st&p=883326
http://www.hydrogenaud.io/forums/index.php...st&p=883151


So, many of us have been here before. I notice that Monty had also tried and failed to instruct Amir in 2012.

The Audibility of Typical Digital Audio Filters in a High-Fidelity Pla

Reply #9
The test procedure, though not ABX, was double blind and would meet TOS 8. I can understand your annoyance at their unjustified attack on ABX, but let's set that aside and concentrate on their results.

The second straw man argument relates to the simulation of "Typical Digital Audio Filters". One might think that one would base any criticism of "Typical Digital Audio Filters" on real world typical digital audio filters, but instead a composite of numerous well-known bad digital audio practices that beginners are taught to avoid  was settled on.
The choice of dither (none at all, or rectangular) appeared to have no impact on the results. The raw results indicate a difference was heard based on the filtering alone. The truncation or bad dither didn't make it easier to hear the difference.

The filter was near-brick wall, near (what would be) Nyquist. The content stayed at 96kHz sampling throughout, so it doesn't really matter if the filter rejected the stop band (though it did) as long as it preserved the pass band (which it did).

I think it's unreasonable for you to suggest that this is somehow a bad filter. Based on the psychoacoustic knowledge that says CD quality audio is good enough, what is wrong with this filter? What is it doing that is audible? (Assuming, for the sake of argument, that the people listening can't hear above 20kHz.) The worst thing I can find is that it appears to be rectangular windowed - but it's something like 75dB down at that point, and a mere 4ms away from the peak, so I don't know how anyone would hear that truncation.

It is a steep filter, but you used to think that brick wall filters, sharp as possible, were wonderful things...
http://www.hydrogenaud.io/forums/index.php...=67619&st=0

Why is it a problem now?

Cheers,
David.

P.S. This requires the krabapple disclaimer: If there was a difference heard in this test (which there was, just about) it was a tiny hard to detect just audible difference in very specific circumstances. It wasn't the "night and day" difference claimed for high resolution audio.

The Audibility of Typical Digital Audio Filters in a High-Fidelity Pla

Reply #10
The test procedure, though not ABX, was double blind and would meet TOS 8. I can understand your annoyance at their unjustified attack on ABX, but let's set that aside and concentrate on their results.

The second straw man argument relates to the simulation of "Typical Digital Audio Filters". One might think that one would base any criticism of "Typical Digital Audio Filters" on real world typical digital audio filters, but instead a composite of numerous well-known bad digital audio practices that beginners are taught to avoid  was settled on.
The choice of dither (none at all, or rectangular) appeared to have no impact on the results. The raw results indicate a difference was heard based on the filtering alone. The truncation or bad dither didn't make it easier to hear the difference.

The filter was near-brick wall, near (what would be) Nyquist. The content stayed at 96kHz sampling throughout, so it doesn't really matter if the filter rejected the stop band (though it did) as long as it preserved the pass band (which it did).


Actually, according to the article,  the content used in the test was at 192 KHz. It didn't stay at 192 KHz because it was played through speakers whose top sample rate is 96 KHz.

"To minimise unknown variations in the signal, we
implemented our test lters at 192 kHz, so that orig-
inal and test signals were presented as 192 kHz 24-
bit PCM format to the system independent of test
condition. Signals were sent to a Meridian 818v2
Reference Audio Core using a high-quality USB ca-
ble, which was connected to a pair of Meridian
DSP7200SE digital loudspeakers."

The spec sheet for the DSP7200's says" "32kHz–96kHz sampling rates at up to 24 bit;"

Quote
I think it's unreasonable for you to suggest that this is somehow a bad filter.


From the above I see no recognition of the technical reasons why I find shortcomings in the filter.

Besides the PDF problem with the dithering, there is also a PSD problem.  There is also a transition band width problem. This has been covered on HA.  For example: Summary of Filter Problems


"
To summarize: the recent Meridian tests sabotaged 16/44 and 16/48 processing by intentionally using

(1) suboptimal dither PDF,
(2) suboptimal dither PSD,
(3) and suboptimal low pass filter transition band width.
"

All fully documented in this thread.

There's a larger and clearer point. The paper is titled "The Audibility of Typical Digital Audio Filters in a High-Fidelity Playback System" but gets all of the most important parameters  of that filter wrong.  Besides the 3 items I listed above, its operational sample rate is wrong.  The digital filters in modern high fidelity playback systems typically run at 8x the sample rate which would be about 320 KHz.

Now if the title of the paper were "The Audibility of off-the-wall Digital Audio Filters that we made up in Matlab for the purpose of pandering to high end audiophiles who we miseducated to hate the CD format in the interest of making more money"  that might be a different matter! ;-)  The paper is not what it pretends to be, even just the title.

Quote
Based on the psychoacoustic knowledge that says CD quality audio is good enough, what is wrong with this filter? What is it doing that is audible? (Assuming, for the sake of argument, that the people listening can't hear above 20kHz.) The worst thing I can find is that it appears to be rectangular windowed - but it's something like 75dB down at that point, and a mere 4ms away from the peak, so I don't know how anyone would hear that truncation.


Asked and answered. Someone even relinked my post covering  the topic in the past day or two.

And in case you are mystified by some of my recent posts, studying this matter has raised my consciousness, to the extent that I may not agree with posts and media that I made in the past. ;-)

I make no claims about infallibility.

The Audibility of Typical Digital Audio Filters in a High-Fidelity Pla

Reply #11
Since we start to repeat arguments from the other thread i still see IM of the metal tweeter more likely as the Meridian pet it is about the filter. Still there was no clear answer that lowpassed music sounds bad. It may sound even better

Is troll-adiposity coming from feederism?
With 24bit music you can listen to silence much louder!

The Audibility of Typical Digital Audio Filters in a High-Fidelity Pla

Reply #12
Since we start to repeat arguments from the other thread i still see IM of the metal tweeter more likely as the Meridian pet it is about the filter. Still there was no clear answer that lowpassed music sounds bad. It may sound even better



I tried to follow up by searching on IM metal dome tweeters, and found no reliable technical data, just a bunch of audiophile opinions based on sighted evaluations.

Can anybody help me with reliable evidence of this alleged problem?

BTW I have a pair of NHT 2.5i that if memory serves have metal dome tweeters in a store room. Their tweeters had nothing to do with their current assignment.

I also just lately replaced the tweeters in a pair of Infinity Primus PC251s whose failed tweeters used some kind of ceramic composite. The replacments were Vifa fabric dome tweeters. Some expert listeners have reported a reduction in nonlinear distortion with sine sweeps, but we think there was a big improvement in Xmax with the new tweeters. Interestingly enough the fabric domes have more HF extension > 25 KHz.


The Audibility of Typical Digital Audio Filters in a High-Fidelity Pla

Reply #14
From the above I see no recognition of the technical reasons why I find shortcomings in the filter.
That's because you haven't explained any reasons.

You keep talking about the bad dither, which is pointless because no one detected it.

You talk about them "sabotaging the tests by intentionally using suboptimal low pass filter transition band width."

You're saying a 500Hz filter transition width is too narrow?
You're also saying it's not representative?

Have you checked what transition width Cool Edit Pro defaults to when resampling? It's about 200Hz at 256 quality, and less than 100Hz at 999 quality. So if you've ever used Cool Edit Pro to resample anything, you've done much "worse" than Meridian's "typical" filter.


Quote
And in case you are mystified by some of my recent posts, studying this matter has raised my consciousness, to the extent that I may not agree with posts and media that I made in the past. ;-)

I make no claims about infallibility.
It's good to know you can change your mind.

What do you now think is the right design for anti-alias and anti-image filters to enable 44.1kHz-sampled audio to be transparent (at least for those who cannot hear above 20kHz)?

Cheers,
David.

The Audibility of Typical Digital Audio Filters in a High-Fidelity Pla

Reply #15
From the above I see no recognition of the technical reasons why I find shortcomings in the filter.
That's because you haven't explained any reasons.


You haven't provided any specific questions for me to answer.

Quote
You keep talking about the bad dither, which is pointless because no one detected it.



That would appear to be a supposition, given that ABX tests don't give specific answers about why the observed results were obtained.

Quote
You talk about them "sabotaging the tests by intentionally using suboptimal low pass filter transition band width."


Yup.

Quote
You're saying a 500Hz filter transition width is too narrow?


If its audible, then of course.

Quote
You're also saying it's not representative?


Asked and answered.

Quote
Have you checked what transition width Cool Edit Pro defaults to when resampling? It's about 200Hz at 256 quality, and less than 100Hz at 999 quality.


Yes. I've even posted similar results.  I appears that there may be some problem with reading comprehension, or maybe just shooting from the hip.

Quote
So if you've ever used Cool Edit Pro to resample anything, you've done much "worse" than Meridian's "typical" filter.


Yes, and I've publicly disavowed some relevant audio test files that I posted,  on exactly those grounds.  You didn't get the memo? ;-)

Quote
Quote

And in case you are mystified by some of my recent posts, studying this matter has raised my consciousness, to the extent that I may not agree with posts and media that I made in the past. ;-)

I make no claims about infallibility.


It's good to know you can change your mind.

What do you now think is the right design for anti-alias and anti-image filters to enable 44.1kHz-sampled audio to be transparent (at least for those who cannot hear above 20kHz)?


Whatever it takes for the most transparent possible results at a reasonable cost. 

Next time I post files for people to ABX re: these issues, the best knowledge I have at the time will be embodied in them.

Here's the nut of the situation. I've seen the Matlab manual for simulating this kind of filter (The one on Octave is similar) and it appears that there are parameters that control these issues.

In that context, how much planning and effort does it take to make the statement: "The Audibility of Typical Digital Audio Filters in a High-Fidelity Players" actually mean something?