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: Audibility of "typical" Digital Filters in a Hi-Fi Playback  (Read 367743 times) previous topic - next topic
0 Members and 21 Guests are viewing this topic.

Audibility of "typical" Digital Filters in a Hi-Fi Playback

Reply #250
There are no obvious flaws.
http://robertwannamaker.com/writings/ieee.pdf
Page 31 figure 5 (b) shows the noise modulation due to rectangular dither.
Page 34 figure 7 shows the lack of noise modulation with triangular dither.

Wannamaker, Lipshitz, and Vanderkooy have spread this information, with varying levels of detail, across the world of audio since the 1980s. I am sure that the authors of the BS paper know and understand this information.

The BS paper accurately reports what they did, and tells you (at least partly) why they intentionally introduced this flaw into the test.

Saying "there are no obvious flaws" is just silly.

Cheers,
David.

Looks like the plot is lost.


More indications of poor reading comprehension. The plots are just where the post says they are and they speak loudly to any person who is even vaguely informed on this topic.

Quote
You are giving me a tutorial on noise modulation and dither.  That has nothing to do with the topic at hand or the question I asked.


Even more indications of poor reading comprehension. The issue being discussed is exactly noise modulation and dither, more specifically the well known (by most of us) adverse effects of rpdf dither on noise modulation and other potentially audible badness.

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.

Audibility of "typical" Digital Filters in a Hi-Fi Playback

Reply #251
Even more indications of poor reading comprehension. The issue being discussed is exactly noise modulation and dither, more specifically the well known (by most of us) adverse effects of rpdf dither on noise modulation and other potentially audible badness.

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.

Those are claims.  This is reality from the paper:



Now explain why you dispute the line I have highlighted in yellow and expressed very clearly on this graph from the paper.

And I like to see you go on record that dither shown at those levels is audible, and a major flaw in any conversion from 24 bits to 16.
Amir
Retired Technology Insider
Founder, AudioScienceReview.com

Audibility of "typical" Digital Filters in a Hi-Fi Playback

Reply #252
It is the same thing in Meridian test but using an advanced methodology (for lay people) using Matlab real-time processing.
"Real-time"?

You need to demonstrate using the paper and data within, how it was that people heard noise modulation and not effect of filtering/dither.
Since noise modulation is the effect of using that incorrect dither, I don't know how to answer that self-contradictory question.


I thought xnor had already explained the other part pretty well. Many signals are sufficiently self-dithering for this to be academic, but with carefully chosen samples and sufficient gain you can demonstrate that the choice of dither can be audible. None of this is news to anyone.

Go on - remind me again how you want to pay extra to get the original bits from the studio to avoid this - because the way to make sure that the rest of the astronomically-more-complex recording chain is OK is obviously to find some guy who you can't even trust to leave the "dither" box ticked, and pay him extra money because he can't be trusted  I mean, honestly, what could possibly go wrong?!

Cheers,
David.

P.S. You said something about the CD version being the dumbed down version for teenage girls. Have you heard the amount of dynamic range compression on some hi-res recordings?! But of course, you have the "original bits", so they must be wonderful?  Not that you are daft enough to believe this, but plenty are.

Audibility of "typical" Digital Filters in a Hi-Fi Playback

Reply #253
Now explain why you dispute the line I have highlighted in yellow and expressed very clearly on this graph from the paper.
Do you understand the significance of the text immediately before the part you highlighted in yellow? "Noise spectral density is derived in a 1-Hz bandwidth"?

I think if I had time to do the calculation properly, I might find there's enough dither noise in a critical band to touch the threshold of hearing. That's the kind of thing that some people might just be able to detect.

Cheers,
David.

Audibility of "typical" Digital Filters in a Hi-Fi Playback

Reply #254
Is amir deliberately obtuse? Is he trolling? Or is that just the way he is?


amir, do you have the tracks used in this experiment? The algorithms used for processing? The software/hardware used to do the AB switching, listening ... ?
"I hear it when I see it."

Audibility of "typical" Digital Filters in a Hi-Fi Playback

Reply #255
You need to demonstrate using the paper and data within, how it was that people heard noise modulation and not effect of filtering/dither.
Since noise modulation is the effect of using that incorrect dither, I don't know how to answer that self-contradictory question.

I thought xnor had already explained the other part pretty well. Many signals are sufficiently self-dithering for this to be academic, but with carefully chosen samples and sufficient gain you can demonstrate that the choice of dither can be audible. None of this is news to anyone.

I post the relevant data on this in my response to Arny.  Please show using the gain in their setup how noise modulation was easily audible.  This is what I meant when I said to back your statements using the report.  You can't critique it in absentia.

Alternatively, there is only one track in this test.  It should be easy then to demonstrate how dither noise modulation caused a major flaw and was readily audible.  Or else it would not be a major flaw, right?

For now, these are the subjective observations we have from the paper from the test subjects:

It was reported that filtering gave "softer edges" to
the instruments, and "softer leading edges" to musi-
cal features with abrupt onsets or changes. Echoes,
when audible, were identi ed as being affected the
most clearly by the ltering. It was felt that some
of the louder passages of the recording were less ag-
gressive after filtering, and that the inner voices (sec-
ond violin and viola) had "a nasal quality". Over-
all, the filtered recording gave a "smaller and
flatter auditory image", and speci cally the physical space
around the quartet seemed smaller.

Listeners described that quantization gave a "rough-
ness" 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 hy-
potheses of Stuart [29] that 16 bits are not sucient
for inaudible quantization.


Would you say these are consistent with audible effects of noise modulation?

Edit: fixed the OCR problems in cut and paste of the quote above.
Amir
Retired Technology Insider
Founder, AudioScienceReview.com

Audibility of "typical" Digital Filters in a Hi-Fi Playback

Reply #256
Is amir deliberately obtuse? Is he trolling? Or is that just the way he is?

More disappointing personal commentary.  Looks like HA forum is much less professional in this manner than I thought prior to joining the forum.

Quote
amir, do you have the tracks used in this experiment? The algorithms used for processing? The software/hardware used to do the AB switching, listening ... ?

I have the tracks.  They are at 96 Khz and I bought them because I liked the music.

I do not have the algorithms.

I do have Matlab.  But don't have the GUI front-end they wrote to perform the test.

How about you?  Do you have the paper and read it? Downloaded the tracks?
Amir
Retired Technology Insider
Founder, AudioScienceReview.com

Audibility of "typical" Digital Filters in a Hi-Fi Playback

Reply #257
Explain how that is an obvious flaw AJ.  Can you do that?

Two words: noise modulation.

And the data that shows it to be in effect in this test is where?

The argument is that rectangular dither may have affected the results.  If you're insisting that the argument has been refuted, which appears to be the case (the insistence, mind you) then, you are essentially claiming that dither was not responsible.  This is the claim that requires supporting evidence.  It's easy enough to test if that was the intention of the authors, so where is the hard data?

I also like to see people go on record and state that anytime a 24 bit file is converted to 16 bit using rectangular dither, the conversion is highly flawed and lossy.

I love the smell of moldy straw in the morning...

Audibility of "typical" Digital Filters in a Hi-Fi Playback

Reply #258
More disappointing personal commentary.

Well, if several people have to point out to you several times basic stuff (such as burden of proof), but you keep repeating nonsense... (which is very disappointing when reading the thread) at some point people will question if you're up to the discussion.

I do not have the algorithms.

I guess it's the same for the hardware. Then why do you expect others to have the software/hardware used, so they could reproduce the experiment accurately (a basal requirement in science, btw) and pinpoint flaws?
"I hear it when I see it."

Audibility of "typical" Digital Filters in a Hi-Fi Playback

Reply #259
Explain how that is an obvious flaw AJ.  Can you do that?

Two words: noise modulation.

And the data that shows it to be in effect in this test is where?

The argument is that rectangular dither may have affected the results.  If you're insisting that the argument has been refuted, which appears to be the case, you are essentially claiming that dither was not responsible.  This is the claim that requires supporting evidence.  It's easy enough to test if that was the intention of the authors, so where is the hard data?

Here it is again:



And the comments from listeners.

Quote
I also like to see people go on record and state that anytime a 24 bit file is converted to 16 bit using rectangular dither, the conversion is highly flawed and lossy.

I love the smell of moldy straw in the morning...

No, it is the stink of not wanting to admit noise modulation at such low levels is audible, yet was an obvious/audible flaw in the test.
Amir
Retired Technology Insider
Founder, AudioScienceReview.com

Audibility of "typical" Digital Filters in a Hi-Fi Playback

Reply #260
More disappointing personal commentary.  Looks like HA forum is much less professional in this manner than I thought prior to joining the forum.

Haha, you've always been a kidder amir. 
Now, regarding (BS) commentary, what do you make of this:

Why does BS use TPDF in their own products for "transparent processing", then use (seemingly) non-transparent RPDF to fabricate desired results?

But don't have the GUI front-end they wrote to perform the test.

Nor the speakers?
So you are purely assuming transparency (wishful thinking fallacy?), since there is no indication in the paper either was tested.
You're choosing to ignore both BS and JJ warnings regarding false positives due to the strong possibility of artifacts with any such test? Why?

cheers,

AJ
Loudspeaker manufacturer

Audibility of "typical" Digital Filters in a Hi-Fi Playback

Reply #261
Yes, and obvious flaws exist which were touched upon, which you have not been able to dismiss successfully.

If dither wasn't a problem, which hasn't been shown to my satisfaction*, then you're still left to contend with differences caused by artifacts in hardware/software; to which you also don't have a satisfactory answer.

So, sure, the thesis *may* have been supported by the results, but with severe caveats.

This isn't particularly earth-shattering when it has already been conceded that 16 bits is not audibly transparent under all circumstances.  I'm pretty sure most here will also accept that not all SRCs will guarantee transparency, either.



Indeed. No one , including Meyre and Moran, has claimed 16bit is never audible. No one has claimed that SRC *cannot* be audible.  No one has said that Redbook is transparent to *all* audible signals.

The issue, for me, is entirely this: since the dawn of hi rez, the reviews have been replete with reports of obvious improvement over Redbook.  You need merely listen to hear the obvious  improvement (if you don't, you're deaf, or 'your equipment is not resolving enough') . Do any factors uncovered in DBT  account for such widespread reports, gleaned from all sorts of different systems, using all sorts of different recordings, and all sorts of listeners, united only in their firm conviction that hi rez 'sounds better'?



Quote
However, this topic is more about what people will attempt to make of the results and the predictable posturing they will take.

Did I get that right, krab?



Partly. I started this thread as a discussion forum for the paper:


Quote
(NB it has been posted once before on HA in the midst of a thread...but perhaps we can use this thread to discuss it, once it is available)



The posturing is just the usual , though I'm glad HA is getting to see the Dancing Man's performance.  Everyone should see it once.

Audibility of "typical" Digital Filters in a Hi-Fi Playback

Reply #262
hard data?

Here it is again:

And the comments from listeners.

"Should" and the interpretation of anecdotes do not constitute hard data.

It is simple enough to test, so where is the evidence that rectangular dither didn't affect the results of the test?

 

Audibility of "typical" Digital Filters in a Hi-Fi Playback

Reply #263
I think we have slammed the door shut on any statement regarding standing of this paper in the eyes of professional AES community.  It is excellent work.  It is the most careful double blind test of small differences due to coding of audio that has been published.  It is an inconvenient truth to be sure for many, but hopefully the search for knowledge and learning trumps that negative emotion.


It is tragic that such high accolades are being heaped on a document that libels and/or ignores well known previous technology (ABX testing, BS1116 recommendations, etc.)

The fact that the published document on the AES web site is apparently mislabeled speaks to excess haste and rush to judgement.



- we don't know how fierce  the competition was for the award

- we don't know who the two reviewers were

- we do know that Amir has dismissed peer review before , when it suited him (re: Meyer & Moran  2007)

So we can be sure he won't  make too much of this convention paper having been peer reviewed for a convention presentation.  I mean, he'd just look silly.

Right?

Audibility of "typical" Digital Filters in a Hi-Fi Playback

Reply #264
Even more indications of poor reading comprehension. The issue being discussed is exactly noise modulation and dither, more specifically the well known (by most of us) adverse effects of rpdf dither on noise modulation and other potentially audible badness.

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.

Those are claims.  This is reality from the paper:



Now explain why you dispute the line I have highlighted in yellow and expressed very clearly on this graph from the paper.

And I like to see you go on record that dither shown at those levels is audible, and a major flaw in any conversion from 24 bits to 16.


Amir, you obviously did not comprehend the part of the paper which clearly said that in a 44 KHz channel, the RPDF dither was approximately 2 times or 6 dB higher.

Footnote 5  page 6:
"Since the total power of RPDF dither in a 16-bit channel is constant, spectral density in a 192-kHz channel is approximately 6 dB lower than in a 44.1-kHz channel."  The graphs shown are obviously for 192 KHz sampling, since they extend a great deal beyond the Nyquist frequency of 44.1 KHz sampling, or 22.05 KHz.  Another win!

You also must not have comprehended the graph and text in the now twice cited Wannamaker/Lipschitz paper (p32):

"The error variance, on the other hand, (for rpdf dither) is clearly signal dependent, so that the noise power in the signal varies with the input. This is sometimes referred to as noise modulation and is
undesirable in audio or video signals."

A bit of friendly advice - just because we call a process Noise Shaping or an undesired signal Quantization Noise doesn't mean that they only apply to noise (random or stochastic signals). Quantization Noise can be caused by coherent and even deterministic signals such as music, and involve pseudorandom noise, in which case it can be deterministic and might even be coherent. Then, its audibility follows the rules for coherent signals which raises the JND threshold considerably.  The cited graph in the Meridian paper does not apply.

I am reminded that people who are ultra-quick to play the amateur/professional card most vigorously are often some the biggest posers around. ;-)

Audibility of "typical" Digital Filters in a Hi-Fi Playback

Reply #265
You can't blame the AES, or the paper's authors, for the frankly embarrassing way that amirm is lauding it about, or the way he's combining the results with some other seriously flawed tests to imply that hi-res audio is now easily and routinely ABXed.

Cheers,
David.


Well, remember what he did with that HA wiki graph 'showing' that 320kbps is not transparent?

It's how he do.

And you know he's not just doing it here.  And he's not the only one doing.

So, just as the hi-end crowd had to contend with Meyer & Moran since 2007 (there' s even a letter about M&M in *this month's*, Stereophile -- very conveniently I would add, since it gives the eds an excuse to tout the Meridian paper*) , now 'we' will have to deal with this, and with parties like Amir flogging it around the Internets like cheerleaders on Red Bull.




*there's also particularly( but not surprisingly) obtuse essay by Art Dudley, on the ever-popular theme of why double blind testing doesn't work for audio.  Which, given that Meridian used DBT in their convention paper, is amusing.

Audibility of "typical" Digital Filters in a Hi-Fi Playback

Reply #266
The posturing is just the usual , though I'm glad HA is getting to see the Dancing Man's performance.

You mean attempt to shift the burden of proof and demand evidence to counter the presentation of data that doesn't support his position?

Or ignore posts that are deserving of a response?

Or argue a different topic?

Or engage in personal attacks?

No, this hasn't gone unnoticed.

Audibility of "typical" Digital Filters in a Hi-Fi Playback

Reply #267
The posturing is just the usual , though I'm glad HA is getting to see the Dancing Man's performance.

You mean attempt to shift the burden of proof and demand evidence to counter the presentation of data that doesn't support his position?

Or ignore posts that are deserving of a response?

Or argue a different topic?

Or engage in personal attacks?

No this, hasn't gone unnoticed.



You left out his  po-faced concern trolling when a response to him get too 'personal'.


It's such a well-rehearsed dance of many steps, I'm guessing it's the work of a lifetime -- I don't want to see any of it neglected.

Audibility of "typical" Digital Filters in a Hi-Fi Playback

Reply #268
Can someone upload a 30 second sample to the uploads section please? Include a loud bit and a quiet bit. Thanks.

Cheers,
David.

Audibility of "typical" Digital Filters in a Hi-Fi Playback

Reply #269
FYI, I passed all of my double blind ABX tests using my laptop and headphones.
You probably think you're getting a hard time here, but you should take it as a compliment that no one has challenged you on this. [...] Anyone else who arrived on HA and made this claim would get a really hard time, due to a long history of drivers, sound cards, and transducers that don't take kindly to ultrasonic content. ABXing hi-res vs 16/44.1 on such crappy equipment is easy, and it's not because hi-res is audibly better.

I can re-open the train wreck discussion if you like, though it will get shut down again if arguments that belong here spill over into there like they did before.

Audibility of "typical" Digital Filters in a Hi-Fi Playback

Reply #270
FYI, I passed all of my double blind ABX tests using my laptop and headphones.
You probably think you're getting a hard time here, but you should take it as a compliment that no one has challenged you on this. [...] Anyone else who arrived on HA and made this claim would get a really hard time, due to a long history of drivers, sound cards, and transducers that don't take kindly to ultrasonic content. ABXing hi-res vs 16/44.1 on such crappy equipment is easy, and it's not because hi-res is audibly better.

I can re-open the train wreck discussion if you like, though it will get shut down again if arguments that belong here spill over into there like they did before.

Two obfuscation contests at the same time? You must enjoy Dancing with the stars.
Loudspeaker manufacturer

Audibility of "typical" Digital Filters in a Hi-Fi Playback

Reply #271
*there's also particularly( but not surprisingly) obtuse essay by Art Dudley, on the ever-popular theme of why double blind testing doesn't work for audio.  Which, given that Meridian used DBT in their convention paper, is amusing.

Yes, I'm not quite sure how the Hi-Rez peddlers can utilize this BS paper.
The scam addicts vehemently reject the notion of blind tests, so how do they hang their hats on this one?
Quite the dilemma, unless of course....

cheers,

AJ
Loudspeaker manufacturer

Audibility of "typical" Digital Filters in a Hi-Fi Playback

Reply #272
Amir, you obviously did not comprehend the part of the paper which clearly said that in a 44 KHz channel, the RPDF dither was approximately 2 times or 6 dB higher.

Footnote 5  page 6:
"Since the total power of RPDF dither in a 16-bit channel is constant, spectral density in a 192-kHz channel is approximately 6 dB lower than in a 44.1-kHz channel."  The graphs shown are obviously for 192 KHz sampling, since they extend a great deal beyond the Nyquist frequency of 44.1 KHz sampling, or 22.05 KHz.  Another win!

Yes, another win for science, and a fail for laymen trying to interpret the same.  The test is run at 192 Khz Arny.  Effect of filtering down to 44.1 Khz simulated within (again, just like your tests of upsampling back the 44.1 to 96 Khz).  As such, the graph properly shows its audibility:



But go ahead and move the green line up 6 db if it makes you feel better.  It is still below the threshold of hearing.

Remember, you will lose a much larger battle if you stay on this path Arny.  That such noise levels is audible in a 105 db SPL playback.  And that they are so audible as to be an obvious flaw in the test.

So keep going at your peril.
Amir
Retired Technology Insider
Founder, AudioScienceReview.com

Audibility of "typical" Digital Filters in a Hi-Fi Playback

Reply #273
Yes, another win for science, and a fail for laymen trying to interpret the same.

We'll hang on to this one for a while.

Audibility of "typical" Digital Filters in a Hi-Fi Playback

Reply #274
Amir, you obviously did not comprehend the part of the paper which clearly said that in a 44 KHz channel, the RPDF dither was approximately 2 times or 6 dB higher.

Footnote 5  page 6:
"Since the total power of RPDF dither in a 16-bit channel is constant, spectral density in a 192-kHz channel is approximately 6 dB lower than in a 44.1-kHz channel."  The graphs shown are obviously for 192 KHz sampling, since they extend a great deal beyond the Nyquist frequency of 44.1 KHz sampling, or 22.05 KHz.  Another win!

Yes, another win for science, and a fail for laymen trying to interpret the same.  The test is run at 192 Khz Arny. 


The fail is yours Amir.  When downsampling with dither per test conditions, the quantization error is generated at the point of downsampling, and therefore distributed over the narrower bandwidth which leads to the higher amplitude.

Quote
But go ahead and move the green line up 6 db if it makes you feel better.  It is still below the threshold of hearing.


Prove it Amir, by drawing in the threshold of hearing for continuous tones.

Finally Amir, you keep forgetting that the reliance on flat PSD dither is one of the serious flaws in the recent Meridian tests. The use of perceptually shaped PSD dither rather spectacularly drops the noise level, particularly in the range of frequencies where the ear is most sensitive.

Prove that you even know what  perceptually shaped PSD dither is Amir, by drawing it in as well.