In Search Of Accurate Sound Reproduction: The Final Word!

[esldude]

41 minutes ago, SoundAndMotion said:

Can you justify saying "ridiculously uncommon steepness"? You may well be right, but if you want to challenge their filter design with that claim, you should back it up. I already told AJ that Jackson et al. should have also justified their choice, IMO. But what is a common steepness? How do you know?

 

I know you understand statistics better than that question. I know you don't like the 95% threshold, but that puts you at odds with many, many highly respected journals. I suspect you present it as 56% vs. 50% to intentionally confuse the issue of statistical significance, hoping for a "gee, 56 isn't that much better than 50" response. But for the single condition that is relevant to my posts: 22050 Hz with no requantization, p=0.012.That even meets your 2 sigma preference.

The rest of your questions can be found by reading the paper. Quick summary: I wasn't talking about requantization, just the filtering, so all your dither questions (although important) don't relate to what I wrote.

 

Maybe you have noticed a bit of a furor over replication issues in the medical field.  I do think most research should function to 3 sigma levels.  5% is 2 sigma.  As it stands people use 5%.   However, 95% confidence with a number of trials that leads to correct identification say 75% of the time (15 of 20 trials) is about a more audible and easily perceived difference than one that requires nearly 200 trials and therefore 56% is at a 95% level.  Doesn't mean the latter is not real, just a lesser issue.  When we barely do meet the level needed along with some questionable and possibly obscuring other differences it catches my attention.  BTW, I don't recall any part of that paper reaching p=.012 I'll have to look back over it (and that doesn't reach 3 sigma levels either p=.0027).

 

Another thing not much talked about is the timing of these tests.  Meridian developed and had the roadmap for MQA filtering all done and working on finalizing it.  Only then, after that point was testing about audibility of lower resolution filtering done.  Seems backwards does it not unless all this was always part of the marketing and justification plan.  So as a result, we have this paper to help people believe the better filtering in MQA is a big plus, based upon specially trained listeners in a highly advanced playback system under extraordinarily good listening conditions after special training able to recognize the filtering correctly 56 % of the time.  So play MQA and everyone hears an immediate revelation.  And no blind test papers showing how MQA fares vs 192/24?  Don't you find that a curious and winding tale of MQA?

 

EDIT on the steepness issue.  Actually it is common for DACs to use half band filters.  They may not be -96 db down at 22,050 hz from 20 khz.  Some are, don't know of any that are down much steeper than that.  Chords use very steep filters that are well down by 22 khz more than -96 db, but they use the whole 2 khz wide transition zone.  You can look at the Jurgen plots Stereophile does on DACs if you don't believe me.  Full cutoff in 460 or 500 hz is much steeper than normal.

[AJ Soundfield]

8 minutes ago, SoundAndMotion said:

Wow, my eyes really are playing tricks on me

I can help you there too

 

Quote

 

On 6/18/2017 at 6:03 PM, CuteStudio said:

441.kHz. No one can build a real filter that works for that

 

 

[esldude]

21 minutes ago, mansr said:

I found a copy of the spec. It doesn't say anything about the D/A conversion.

 

 

Okay, thanks for finding that info for us.

[SoundAndMotion]

9 minutes ago, esldude said:

Maybe you have noticed a bit of a furor over replication issues in the medical field.  I do think most research should function to 3 sigma levels.  5% is 2 sigma.  As it stands people use 5%.   However, 95% confidence with a number of trials that leads to correct identification say 75% of the time (15 of 20 trials) is about a more audible and easily perceived difference than one that requires nearly 200 trials and therefore 56% is at a 95% level.  Doesn't mean the latter is not real, just a lesser issue.  When we barely do meet the level needed along with some questionable and possibly obscuring other differences it catches my attention.  BTW, I don't recall any part of that paper reaching p=.012 I'll have to look back over it (and that doesn't reach 3 sigma levels either p=.0027).

 

You are right about 2 and 3 sigma. P=0.012 is table 2, condition 1.

 

9 minutes ago, esldude said:

 

Another thing not much talked about is the timing of these tests.  Meridian developed and had the roadmap for MQA filtering all done and working on finalizing it.  Only then, after that point was testing about audibility of lower resolution filtering done.  Seems backwards does it not unless all this was always part of the marketing and justification plan.  So as a result, we have this paper to help people believe the better filtering in MQA is a big plus, based upon specially trained listeners in a highly advanced playback system under extraordinarily good listening conditions after special training able to recognize the filtering correctly 56 % of the time.  So play MQA and everyone hears an immediate revelation.  And no blind test papers showing how MQA fares vs 192/24?  Don't you find that a curious and winding tale of MQA?

 

All good points. So rather than use guilt by association (with MQA, profit), I think you must find the flaws that render the conclusions implausible. You and AJ bring up many of the types of issues we found in journal club, when we found flaws in most papers. There are no perfect papers, and one has to judge whether the methods call the results into question. Such rigor is great! But you realize that many of the studies Jackson-critics love (e.g. M&M) would simply wither in the face of your level of scrutiny.

 

I'm not a giant fanboy of Jackson, but I find the results compelling and worth citing. At the very least, when CuteStudio calls into question the "correctness" of such filters, then the usual yardstick of "transparency" cannot be used to dismiss the concerns as ridiculous. Yes, "specially trained listeners in a highly advanced playback system under extraordinarily good listening conditions after special training able to recognize the filtering", but it is not transparent. Many people here may feel their experiences and setups are similar.

 

What is your take on the 4-minute-mile? The average human, the typical human, can't do it. But some can. Should we argue "humans can't run a 4-minute-mile"?

[CuteStudio]

16 minutes ago, AJ Soundfield said:

I can help you there too

 

 

So AJ, have you found the filter yet?

Just post up the schematic, I don't mind if it's a long URL or a screenshot, but it would be nice if one didn't have to pay $30 to see it behind a paywall. For a 96k or 192k I suspect a basic RC would probably do a reasonable job, I'm not convinced that holds for 44.1, to my knowledge no one has created a perfect analog filter that does that but I'm waiting to be shown how mistaken I am.

 

It's funny but no-one who says "16/44.1 is perfect" ever seems to use it, have you noticed? They use upsampling and dither. Dither of course is a whole different can of worms and while it must be an intrinsic part of the correct downgrading to 16bit, surely if different dither techniques are audible that's a type of logical flaw in the argument that 16bits is enough? So we now have two uncontrolled processes (upsampling and dither) that appear to be band-aids to make 16bits acceptable.

 

So I guess the reality is more like "16/44.1 is perfect as long as you upsample and chose a good dither, and accept the temporal limitations of dithering".

Which may not really be the same thing at all as being perfect I suppose.

[Ralf11]

This might be useful:

 

CD Quality not quite as good as high bit rate/depth
Reiss, J. D.  2016.  A Meta-Analysis of High Resolution Audio Perceptual Evaluation.  JAES 64(6): 364-379.  

 

http://www.aes.org/e-lib/browse.cfm?elib=18296

[AJ Soundfield]

1 hour ago, CuteStudio said:

So AJ, have you found the filter yet?

How about we cut out the middle man and you ask SAM directly, how Jackson et al created a fake 44.1 filter with MATLAB in the paper he cited. I think direct conversation between you two would be much better.

[fas42]

5 hours ago, esldude said:

But, but, but, but, but,but........this is Frank.  Quality was invariably better.  Frank could hear if Windows had resampled his gigasample music even without being able to hear the speaker with his ear next to it.  This is Frank.  He can show you all the same wonders of music and sound as long as showing you doesn't entail showing you. 

 

Dennis, it's nice to have you around to give me a warm and cuddly feeling  ... I'm sorry my approach doesn't compute for you - understanding that concerning oneself with a whole myriad of little details, as well as the large ones, is key doesn't sit easily with a lot of people; but that's what my message will be, over and over again.

 

It may help you understand more if you're aware that I wrote software for many, many years - most people are happy to push out something with a few bugs which gets the job done, and looks the goods; I was never happy until I was convinced there were zero, I repeat, zero bugs - so, where's the "secret sauce" there ... ?

[fas42]

3 hours ago, CuteStudio said:

It's funny but no-one who says "16/44.1 is perfect" ever seems to use it, have you noticed? They use upsampling and dither. Dither of course is a whole different can of worms and while it must be an intrinsic part of the correct downgrading to 16bit, surely if different dither techniques are audible that's a type of logical flaw in the argument that 16bits is enough? So we now have two uncontrolled processes (upsampling and dither) that appear to be band-aids to make 16bits acceptable.

 

So I guess the reality is more like "16/44.1 is perfect as long as you upsample and chose a good dither, and accept the temporal limitations of dithering".

Which may not really be the same thing at all as being perfect I suppose.

 

Again, "16/44.1 is perfect" - that is, a musical recording that is encoded using that standard provides all the data required to achieve SQ as fine as one could wish for. How a particular system, individual chooses to process that data to achieve the required quality is their choice - the point being, that there is nothing intrinsically missing or 'wrong' with what a playback system finds with a 16/44.1 recording, that prevents the listener getting the full measure of what was recorded.

[mansr]

4 hours ago, CuteStudio said:

So AJ, have you found the filter yet?

Just post up the schematic

Here's the full schematic of the Sony D-50 CD player. The lowpass filter is clearly marked in the top right corner.

[AJ Soundfield]

4 hours ago, CuteStudio said:

It's funny but no-one who says "16/44.1 is perfect" 

Can you quote someone other than you (or crazy Frank) saying that?

 

Oh, btw, folks who advocate for 16/44 home playback being sufficient don't make that specific claim either

[AJ Soundfield]

17 minutes ago, mansr said:

Here's the full schematic of the Sony D-50 CD player. The lowpass filter is clearly marked in the top right corner.

Its not real, no one can build a 44.1k filter

[mansr]

1 hour ago, AJ Soundfield said:

Its not real, no one can build a 44.1k filter

And here's the measured frequency response of that CD player:

[Ralf11]

that graph was rejected by my pet dolphin

[AJ Soundfield]

That D-50 graph is a hoax.

 

Ok SAM, times up. http://journal.frontiersin.org/article/10.3389/fpsyg.2017.00093/full

Quote

High-resolution audio has a higher sampling frequency and a greater bit depth than conventional low-resolution audio such as compact disks. The higher sampling frequency enables inaudible sound components (above 20 kHz) that are cut off in low-resolution audio to be reproduced. Previous studies of high-resolution audio have mainly focused on the effect of such high-frequency components. It is known that alpha-band power in a human electroencephalogram (EEG) is larger when the inaudible high-frequency components are present than when they are absent. Traditionally, alpha-band EEG activity has been associated with arousal level. However, no previous studies have explored whether sound sources with high-frequency components affect the arousal level of listeners. The present study examined this possibility by having 22 participants listen to two types of a 400-s musical excerpt of French Suite No. 5 by J. S. Bach (on cembalo, 24-bit quantization, 192 kHz A/D sampling), with or without inaudible high-frequency components, while performing a visual vigilance task. High-alpha (10.5–13 Hz) and low-beta (13–20 Hz) EEG powers were larger for the excerpt with high-frequency components than for the excerpt without them. Reaction times and error rates did not change during the task and were not different between the excerpts. The amplitude of the P3 component elicited by target stimuli in the vigilance task increased in the second half of the listening period for the excerpt with high-frequency components, whereas no such P3 amplitude change was observed for the other excerpt without them. The participants did not distinguish between these excerpts in terms of sound quality. Only a subjective rating of inactive pleasantness after listening was higher for the excerpt with high-frequency components than for the other excerpt. The present study shows that high-resolution audio that retains high-frequency components has an advantage over similar and indistinguishable digital sound sources in which such components are artificially cut off, suggesting that high-resolution audio with inaudible high-frequency components induces a relaxed attentional state without conscious awareness.

 

The higher sampling frequency enables higher frequency sound components to be reproduced, because one-half of the sampling frequency defines the upper limit of reproducible frequencies (as dictated by the Nyquist–Shannon sampling theorem). However, in conventional digital audio, sampling frequency is usually restrained so that sounds above 20 kHz are cut off in order to reduce file sizes for convenience. This reduction is based on the knowledge that sounds above 20 kHz do not influence sound quality ratings (Muraoka et al., 1981) and do not appear to produce evoked brain magnetic field responses (Fujioka et al., 2002).

That was the citation by Bob Stuart in support of MQA.

Ok, now onto that inaudibility:

Quote

"An astonishingly short two years after writing that, I can report that digital audio has taken a significant step forward in its inexorable march toward superiority over analog. The development to which I refer is called High Definition (HDCD)."

" But, as you might also expect, the 24/192 two-channel (DVD-A) tracks sounded by far the best, and quite significantly so. Everything at 48kHz and below sounded pleasant if not terribly detailed, but when shifting into high gear at 88.2kHz, the resolution became transparent enough to hear the warts in the recording, and even perhaps the limitations of the hardware. And it sounded more liquid, as did all the high-resolution formats."

"Every Stereophile writer who has auditioned DSD under critical conditions—Robert Harley, Peter van Willenswaard, Jonathan Scull, and me—has found it both very much better than 16/44k1 CD and much closer to the analog experience."

"As compelling as the untreated hi-res file sounded, I literally laughed at the difference when the MQA version began. Not only did it feel as though a veil had been lifted, with far more color to the sound, but instruments also possessed more body. With more meat on dem bones, I also noticed less of a digital edge on the violin. I've heard Hahn in concert several times, and this was the closest to real I've ever heard her violin sound on recording."

More "relaxed" eh?

[esldude]

2 hours ago, mansr said:

And here's the measured frequency response of that CD player:

You are not fooling anyone.  Lots of people can make graphs any shape they wish. Nope not buying it.

[sdolezalek]

Buried in the back and forth of this thread are a lot of questions about not only room response, but whether eliminating or minimizing back waves and rear reflections from speakers is a good thing.  As a fan of the forward and backward radiating Magnepans, I know I belong in the camp of those that prefer the sound of that very broad radiating pattern.  I also understand why that would actually increase the amount of "extra" reflected sound in my room, which my brain somehow has to evaluate against the direct waves from each speaker that reach my ears first.  Perhaps the cross cancellations of all those surrounding waves reduce the interference of the lobing that stereo images otherwise induce.  But do we more broadly understand why some of us prefer this "milti-reflectional reality" versus those who prefer to maximize, as much as possible, the purity of the direct wave and eliminate all other?

 

I logically buy into the notion that two forward firing, narrow beam speakers in an anechoic room would present our ears with the closest thing to what was recorded, and I can appreciate the sheer accuracy of that sound, but I have never found it to be a good representation of the real thing.  In fact, the one thing I notice about my Magnepans is that the sound they play sounds more like a real instrument (from an adjacent room) than virtually any box speaker.  So they must be doing something to "load" the room with sound pressure in a way that direct firing monitors do not.  

 

Given how much progress we have made in just a few years with regard to squeezing so much more out of a 16/44 format that most of us thought possible, I wonder how much more we can squeeze out of speakers if we start applying more and better data analytics to what is really going on with loudspeakers and how we hear them? 

[STC]

2 minutes ago, sdolezalek said:

But do we more broadly understand why some of us prefer this "milti-reflectional reality" versus those who prefer to maximize, as much as possible, the purity of the direct wave and eliminate all other?

 

The only way you can listen to the purity of direct waves is when you turn your room to anechoic chamber. 

[sdolezalek]

2 hours ago, AJ Soundfield said:

Ok SAM, times up. http://journal.frontiersin.org/article/10.3389/fpsyg.2017.00093/full

That was the citation by Bob Stuart in support of MQA.

Ok, now onto that inaudibility:

More "relaxed" eh?

 

I guess the debate over whether "hi-res" is better because it a) conveys greater than 20kHz signals or b) simply permits filtering methods in the D/A conversion that result in a clearer 20-20kHz sound, will never end.  Who knows, maybe it is both.  But it is an example of where we can clearly measure that there are differences, either in the frequencies being delivered, or in the impulse response (pre/post-ringing), or frequiency response between hi-res and 16/44 material, but those measurements fail to tell us which of those differences actually cause us, as humans, to legitimately prefer one outcome over the other.  

[sdolezalek]

2 minutes ago, STC said:

 

The only way you can listen to the purity of direct waves is when you turn your room to anechoic chamber. 

I can get pretty close by putting a set of box speakers outdoors and there are thus no side or front or rear walls.  Now, that isn't a sound i particularly like, but my guess is that part of the reason for that is that those speakers were voiced to be used in an enclosed room and would have needed to be given a different response curve if the seller knew that there were going to be used in an anechoic environment.  By the way, I also don't like the Magnepans in that environment (even though they sound quite good if I eliminate both side and rear walls.  

[fas42]

12 minutes ago, sdolezalek said:

I logically buy into the notion that two forward firing, narrow beam speakers in an anechoic room would present our ears with the closest thing to what was recorded, and I can appreciate the sheer accuracy of that sound, but I have never found it to be a good representation of the real thing.  In fact, the one thing I notice about my Magnepans is that the sound they play sounds more like a real instrument (from an adjacent room) than virtually any box speaker.  So they must be doing something to "load" the room with sound pressure in a way that direct firing monitors do not.  

 

Given how much progress we have made in just a few years with regard to squeezing so much more out of a 16/44 format that most of us thought possible, I wonder how much more we can squeeze out of speakers if we start applying more and better data analytics to what is really going on with loudspeakers and how we hear them? 

 

There's a lot of "depends" in there ... trust me , direct firing monitors can do the "sounds more like a real instrument (from an adjacent room)". What the straightforward, box monitors usually lack is a good enough electronics chain driving them, and the stabilising of the speakers in their position is not good enough - I intend to get my cheap, Sharp speakers running again, and will record some clips hopefully demonstrating that quality. Yes, I recently heard Magnepans do good sound, which faded when they changed the source configuration - another, "depends".

 

What we can do for speakers is drive them with a better class of signal, make sure they are locked down as rigidly as possible, and eliminate the silly little weaknesses in many of them.

 

[esldude]

8 hours ago, SoundAndMotion said:

 

You are right about 2 and 3 sigma. P=0.012 is table 2, condition 1.

 

 

All good points. So rather than use guilt by association (with MQA, profit), I think you must find the flaws that render the conclusions implausible. You and AJ bring up many of the types of issues we found in journal club, when we found flaws in most papers. There are no perfect papers, and one has to judge whether the methods call the results into question. Such rigor is great! But you realize that many of the studies Jackson-critics love (e.g. M&M) would simply wither in the face of your level of scrutiny.

 

I'm not a giant fanboy of Jackson, but I find the results compelling and worth citing. At the very least, when CuteStudio calls into question the "correctness" of such filters, then the usual yardstick of "transparency" cannot be used to dismiss the concerns as ridiculous. Yes, "specially trained listeners in a highly advanced playback system under extraordinarily good listening conditions after special training able to recognize the filtering", but it is not transparent. Many people here may feel their experiences and setups are similar.

 

What is your take on the 4-minute-mile? The average human, the typical human, can't do it. But some can. Should we argue "humans can't run a 4-minute-mile"?

 

I have no argument with the results as they are.  The extra steep filtering yes.  Why?  Because if we relax the filter that was just barely detected to 1000 hz transition does the audibility of it go away?  I don't know, it seems a likely conclusion it might go away at some point.  It is telling that no testing at a 2050 hz transition was done.  Suppose that indicated the issue disappears? 

 

Yes the M&M paper has some issues too.  No question. 

 

There are people who knowing how human hearing worked think the small transition zone might in some small number of people be audible.  JJ Johnston among them.  He thinks the 4 khz transition with 48 khz sampling is almost enough to dismiss the possible problem, but a bit more to be certain.  96 khz is more than enough in his view.

 

So sure, I might test a 100 people and not get a 4 minute miler.  Yet real they are.  If you want to compare this to such outliers, I have no issue with it.  Should we completely alter the audio world of music formats for the audiophile equivalent of 4 minute milers?  If so what would be required? Raising the sample rate to 65 khz and start using a transition zone of 7500 hz would do it.  Or simply using existing 88 khz or 96 khz rates would do it.   Pretending that either of those would benefit most audiophiles would be wishful thinking or maybe sensible overkill for certainty.  Thinking it would be a revelatory improvement to anyone with music is fantasy.

 

 

[fas42]

16 minutes ago, sdolezalek said:

 

I guess the debate over whether "hi-res" is better because it a) conveys greater than 20kHz signals or b) simply permits filtering methods in the D/A conversion that result in a clearer 20-20kHz sound, will never end.  Who knows, maybe it is both. 

 

b) is the answer, always. Having downloaded hi-res files and then looked at what the >20kHz content was, it's stuff like the performer accidently knocking something, or some very low level material in the midst of a massive crescendo. 100% discardable, IOW ...

[STC]

5 minutes ago, sdolezalek said:

I can get pretty close by putting a set of box speakers outdoors and there are thus no side or front or rear walls.  Now, that isn't a sound i particularly like, but my guess is that part of the reason for that is that those speakers were voiced to be used in an enclosed room and would have needed to be given a different response curve if the seller knew that there were going to be used in an anechoic environment.  By the way, I also don't like the Magnepans in that environment (even though they sound quite good if I eliminate both side and rear walls.  

 

You need lateral reflections for naturalness. Otherwise the sound from the speakers will sound too clinical and dead. It doesn't matter what frequency response the designer chooses. 

 

Here is a good read

 

http://www.pnas.org/content/111/12/4409.full.pdf

 

[SoundAndMotion]

5 hours ago, AJ Soundfield said:

Ok SAM, times up. http://journal.frontiersin.org/article/10.3389/fpsyg.2017.00093/full

That was the citation by Bob Stuart in support of MQA.

Ok, now onto that inaudibility:

More "relaxed" eh?

LOL. Did the easter bunny bring you an egg timer? I don't schedule my life around your posts. Your dog whistle is inaudible to me... LOL again.

 

I see my name, but I don't see anything in your post for me!? I didn't say anything about audibility of >20kHz, and I'm not obsessed with Bob Stuart, as you seem to be. Did you mistake me for someone else? Or was there a glitch in your reading comprehension... or, wait... I know your smart AJ, so I'm betting on you intentionally knocking down your straw man, which doesn't interest me.

 

By the way, citing an article does not mean supporting everything in it. You are very focussed on a few adjectives (inaudible, etc.), but I suspect Stuart cited the article for the data, not every word therein.