Skip to main content

Topic: Audibility of "typical" Digital Filters in a Hi-Fi Playback  (Read 78844 times) previous topic - next topic

0 Members and 1 Guest are viewing this topic.
Audibility of "typical" Digital Filters in a Hi-Fi Playback
Reply #425
So, for the love of the flying spaghetti monster, could you please finally answer these simple questions amirm?


Day one of "Dealing with Amir" 101: The clearer, the more revealtory the question, the less likely you are to receive a clear answer. They don't call him "The Dancing Man" for nothing! ;-)

IME this sort of behavior can get you into surprisingly high places in business, but usually it doesn't last forever.

  • ajinfla
  • [*][*][*][*][*]
Audibility of "typical" Digital Filters in a Hi-Fi Playback
Reply #426
No, I started up pretty "deaf."  I remember being shocked that I could not hear much difference between 128 kbps MP3 and the source.  That bothered me .  So I started on a path to become a trained listener and after literally hundreds of hours of testing, and tons of education learning about psychoacoustics and algorithms in audio processing, all of a sudden found myself in a super unique situation of outperforming everyone else around me in such tests.  I have never liked the phrase but everyone would call me the "golden ear."

And let's distinguish between hearing and listening.  My hearing is shot due to age.  I am embarrassed to say that I can't even hear 12 Khz well.  The reason I do well still I think is that I focus on what I think technically matter and combine that with my training to look for small differences.  I suspect people with my training but intact frequency response will be able to do a lot better.  I actually ran into one such person at Microsoft who was working with a partner of ours.  He was hearing high frequency distortions I could not.  We hired him immediately .  And I moved off from doing a lot of the listening tests myself.

Let me confess.  I had no idea I could do this after so many years after retirement. I just ran the tests because people like Arny kept egging me on so I gave it a try.


Good morning amir,

Something about this little story has been bugging me a bit, perhaps you could enlighten us. 

First, let me state, that if I could outrun Usain Bolt, the entire world would know about it. For obvious reasons, if I claimed to be able to do so, unsupervised, only in my backyard, by myself, no witnesses or oversight, folks might rightly be a bit skeptical and want to see a demo, maybe at the Olympics or some other similar witnessed, supervised, documented event, on camera, with electronic timing, etc, etc, etc.

Ok, so you claim above to have become this elite aural athlete, a real Golden Ear (per your coworkers). Seemingly many, many years ago (prior to retirement), you could "listen" to 16/44 (which had been around for a lot longer) and hear artifacts.
This certainly would predate M&Ms 2007 myth-buster.
Is there any record anywhere, of you demonstrating this elite aural athleticism, at say, the fully documented listening Olympics, supervised, to your peers? Or anyone (Obviously there was no mention in the +/-10% volume debacle on AVS in 2009)?
Or is the very first demo to the world earlier this year (2014), when you the ex-MS exec, sat at your Windows pc (possibly with your occasionally barking dog as the sole witness) and "passed" Arnie's corrupt ABX windows computer file ABX online test? And all similar subsequent ones of course.
Just curious, is all. Thanks. 

cheers,

AJ
Loudspeaker manufacturer

  • amirm
  • [*][*][*][*]
Audibility of "typical" Digital Filters in a Hi-Fi Playback
Reply #427
OK, finally I get to this technical point:

[...]
Having read the Stuart paper and prior citations, I am starting to think this may be due to time domain impact.  It certainly is not frequency domain since I can't hear the ultrasonics. Just as pre-echo is a very audible time domain artifact (although created in frequency domain), maybe these filters act the same way.


After reading this I get an even stronger sense that you must have been cheating in a lot of the logs you posted, or you have some seriously anomalous hearing (regarding detecting lossy compression).

The filters used have linear phase. The pre-echo is no echo but filter ringing at over 21 kHz that only acts on the energy that is up there.
If you can't hear well past 12 kHz then how would you detect the ringing of a filter at over 21 kHz? So please explain why this filter ringing is of concern for you.

The answer is right in what I wrote but since it was misunderstood, I am going to explain the basics.

What you describe is not "ringing" but passband ripple.  Yes, it is sometimes called ringing but given the fact that I was talking about time domain (see the new highlight in red), you shouldn't have gone to frequency domain.

There is no disagreement there.  The paper makes it clear and I have said the same that the filters are near perfect in frequency domain:

For both FIR lters, the ripple depth over the passband was a maximum of 0.025 dB, and the stopband attenuation was 90 dB.

Note the correct terminology of "ripple" not ringing.  But yes, that is as ruler flat as we need it with just +-.025db variation and aliasing truncated by 90 db.  I wouldn't be able to hear that 0.025 db variation if it were in the range that I can hear.  And this as you say is in the range that I can't hear.  So again, no disagreement that that the frequency domain analysis doesn't indicate why this could be audible.

What followed in my post was a hypothesis.  It is a hypothesis that is stated in the paper as I indicated.  As any hypothesis, it is being offered as a potential answer, not proof.  There is foundation in it but not one that I have personally investigated and hence the phrase I used: "I am starting to think..."

What is that thinking?  It is what I described.  What is happening in time domain.  I created two filters in Matlab using the same 0.025 db passband ripple and 90 db out of band rejection.  The first one is very similar to what Stuart used in his study (not identical because I am not performing any optimization):

.

The way you refer to pre-echo as "not being an echo" is totally nonsensical.  Of course it is not an echo.  It is *pre*-echo as it happens prior to the signal itself.  From our old friend the wiki: http://en.wikipedia.org/wiki/Pre-echo

Pre-echo (not to be confused with reverse echo) is a digital audio compression artifact where a sound is heard before it occurs (hence the name). It is most noticeable in impulsive sounds from percussion instruments such as castanets or cymbals.

Hey, what do you know?  It says the same thing I did .

The psychoacoustic component of the effect is that one hears only the echo preceding the transient, not the one following – because this latter is drowned out by the transient. Formally, forward temporal masking is much stronger than backwards temporal masking, hence one hears a pre-echo, but no post-echo.

It did it again! 

So as you see, the whole confusion is due to thinking we are talking about filter ripple and frequency domain, when in multiple references it was clear that I was talking about time domain.  And hence, my hearing limitation in frequency domain is no barrier to hearing such a phenomena.  Learning to hear pre-echo even in minute amounts is a skill I had to develop to hear compression artifacts.  So if that is what is at play, then that is the reason I may be hearing the difference.

Stuart makes multiple references to the same thing:

These parameters were chosen to o er a reasonable
match to the downsampling lters used in good-
quality A/D converters or in the mastering process;
we wanted to minimise the ripple depth and max-
imise the stop band attenuation in order to reduce
audible ringing artefacts, as described by Lagadec
[31].


Have you read Lagadec's paper?  If not, I suggest you do so.  It explains the above and includes with it (informal) listening test observations of the impact of such ringing.

When the analysis was restricted to just the high-
yield audio sections, performance was signi cantly
better for the 48-kHz lter than for the 44.1-kHz l-
ter. This is perhaps surprising given that the difference
in spectral content between the filtered signals
was between 22050 Hz and 24000 Hz. A time-domain
explanation could be that the length of the 44.1-
kHz lter was longer than the 48-kHz one: 4.25 ms
compared with 3.9 ms; this could have resulted in
350 s less pre- and post-ringing, potentially render-
ing the 44.1-kHz [sic I think he means 48 khz] filter less audible.
This explanation is consistent with the idea of time smearing of ne
temporal details mentioned earlier [1, 14].


So now you see why I needed a bit of time to respond.

The above is also the reason I say that if CD had picked 48 Khz as the sampling, I would have no beef with it.  We have plenty of room to implement our filter above 20 Khz. But by picking 44.1, it leaves us a small margin forcing sharper filters and more time domain ringing.

Amir
Retired Technology Insider

  • 2Bdecided
  • [*][*][*][*][*]
  • Developer
Audibility of "typical" Digital Filters in a Hi-Fi Playback
Reply #428
Applying the same analysis to the sharp cut-off filter that was tested in the listening test gives us 9.5-2.5 = 7 milliseconds.  In other words the span of ringing is now about 5 times wider in time than it was in the short cut off.

Is this audible?  I gave a reference in my post as to how it might be: pre-echo.  In psychoacoustics there is a concept of temporal masking.  If you hear a loud sound, what follows it very quickly may not be audible.  So if we look at our impulse itself, it creates a pretty nice shadow over what is to follow it.  But looking behind it, there is impulse (assuming silence for now) and backward temporal masking is quite limited.  This could, could, make the pre-ringing audible.

Note that since this is in time domain, it has nothing to do with my ability to hear high frequencies.
That bit doesn't make any sense. The ringing is at the filter cut-off frequency: 24kHz. If you can't hear 24kHz, you can't hear 24kHz. The fast it's before an impulse, or fades in, doesn't help you.

Quote
All audio samples go through the above transformation, not just high frequency ones.  My truncated frequency response has no bearing on this situation unless the test tones are only above my hearing which is not the case in this musical segment.  I am pretty sure I can hear what music is playing .
Your truncated frequency response matters because the pre-echo is always 24kHz, and the frequencies in the musical segment matter, because unless they get near 24kHz (i.e. have signal components or transitions or tones or clicks that reach up that far) they don't excite it. You won't see it.


It's not to say I reject this theory completely. It's been advocated many times. If there's something a little wrong anywhere in the system that makes ultrasonic signals fall down into the audible band in some way, then less pre-ringing could be audibly better, even when it's originally only at 24kHz.

But you need something that makes the 24kHz signal drop into the audible range before you can start talking about temporal (un)masking. Most of the things that could do that are things you don't want in a decent audible system.

Cheers,
David.

P.S. You've got 3kHz between 19kHz and 22kHz with CD's 44.1kHz sampling. You can have your gentle filter. As I'm sure you know, Bob will sell you one on the playback side. It'll cut so early (18kHz IIRC) and gently that it will removing all that nasty 22kHz ringing, even if it's in the source.

  • amirm
  • [*][*][*][*]
Audibility of "typical" Digital Filters in a Hi-Fi Playback
Reply #429
As I'm sure you know, Bob will sell you one on the playback side. It'll cut so early (18kHz IIRC) and gently that it will removing all that nasty 22kHz ringing, even if it's in the source.

I have a much easier solution that doesn't require new hardware from anyone: get the high sampling rate version of the content.  Problem solved.  You all can do the same if you choose.  Tell me again why you guys are arguing.
Amir
Retired Technology Insider

  • 2Bdecided
  • [*][*][*][*][*]
  • Developer
Audibility of "typical" Digital Filters in a Hi-Fi Playback
Reply #430
Tell me again why you guys are arguing.
Again?! Isn't 18 pages of it sufficient?

  • greynol
  • [*][*][*][*][*]
  • Global Moderator
Audibility of "typical" Digital Filters in a Hi-Fi Playback
Reply #431
I have a much easier solution that doesn't require new hardware from anyone: get the high sampling rate version of the content.  Problem solved.  You all can do the same if you choose.  Tell me again why you guys are arguing.
The false puffiness over the fear that SRCs will cause audible artifacts in 5 of 9 attempts under specific criteria, when far larger issues remain unresolved in the process as they relate to what is actually provided to the buying public?
  • Last Edit: 21 November, 2014, 12:30:04 PM by greynol
Is 24-bit/192kHz good enough for your lo-fi vinyl, or do you need 32/384?

Audibility of "typical" Digital Filters in a Hi-Fi Playback
Reply #432
P.S. You've got 3kHz between 19kHz and 22kHz with CD's 44.1kHz sampling. You can have your gentle filter. As I'm sure you know, Bob will sell you one on the playback side. It'll cut so early (18kHz IIRC) and gently that it will removing all that nasty 22kHz ringing, even if it's in the source.


It has already been pointed out on this thread that real world CD players using actual real world DAC chips have low pass filter transition bands that are 2-3 KHz wide, and not the 0.5 KHz transition bandwidth that Meridian is trying to tell us characterize a "Typical Digital Filter in a HiFi Playback system".  You can count on amateur experimenters to not catch this detail or understand its significance.

Whatever wave files that amateur listeners are actually listening to and with what options is mystery meat since they will still use the old Foobar Release 1.x  ABX comparator, prepare their own test files, and won't put them up for independent examination.

Amateurs also often fail to use standard terminology while characterizing ringing which generally involves 10% variations - 10% below peak (90%) to a 10% or less residual.

Audibility of "typical" Digital Filters in a Hi-Fi Playback
Reply #433
The above is also the reason I say that if CD had picked 48 Khz as the sampling, I would have no beef with it.  We have plenty of room to implement our filter above 20 Khz. But by picking 44.1, it leaves us a small margin forcing sharper filters and more time domain ringing.
Glad to see we're back to filters
I finally got the (interesting) paper and am still in the process of digestion. Plenty of questions.
First of all I'd like to compliment the authors with an unprecedented promotion of Haydn string quartets!
The paper doesn't IMO really separate the issues of dither and filtering. Filtering alone would be Conditions 1 and 4, although I can't find info how the processed signal was converted to 24 bits to feed the monitors (truncation, dithering?).
Am I the only one who thinks that a 90 dB stopband attenuation isn't spectacular ? Perhaps it's good enough, but when testing for very low level artifacts almost anything might matter. What about the <15 dB noise of the speakers (spec from website)? It would have been nice to see its spectrum in Fig.3.

About pre-ringing: when comparing a hi-res signal and a filtered (SRC) version in a null-test, there is no significant difference (frequency and time domain) in the passband. I would like to see more evidence that the >20kHz ringing is audible. I remain skeptical.

In 4.4 the authors announce future tests which include "TPDF dither with and without noise-shaping filters such as those used to limit noise in DSD playback, minimum-phase filters and apodizing filters". Since the processing is done in Matlab, they could have done those tests right away, IMHO. I'd like to see tests that not only determine which formats are not good enough, but also investigate which format can be really transparent (excluding abuse by incompetent engineers). As a recording engineer I'd like to know if 24/96 is good enough, or if there is evidence that we need to go even higher.

  • Wombat
  • [*][*][*][*][*]
Audibility of "typical" Digital Filters in a Hi-Fi Playback
Reply #434
Here is an older pic i used for something different but it may show where the terrific ringing happens with different filters.
Is troll-adiposity coming from feederism?
With 24bit music you can listen to silence much louder!

  • amirm
  • [*][*][*][*]
Audibility of "typical" Digital Filters in a Hi-Fi Playback
Reply #435
Tell me again why you guys are arguing.
Again?! Isn't 18 pages of it sufficient?

I would think so and I am looking at ending my participation.  It is a shame though that I can't get people to agree with the simple business principal that high resolution downloads are happening.  And by downloading such bits I am free of worrying whether downgrading them is audible or not.
Amir
Retired Technology Insider

  • amirm
  • [*][*][*][*]
Audibility of "typical" Digital Filters in a Hi-Fi Playback
Reply #436
I have a much easier solution that doesn't require new hardware from anyone: get the high sampling rate version of the content.  Problem solved.  You all can do the same if you choose.  Tell me again why you guys are arguing.
The false puffiness over the fear that SRCs will cause audible artifacts in 5 of 9 attempts under specific criteria, when far larger issues remain unresolved in the process as they relate to what is actually provided to the buying public?

I have no fear.  I am simply avoiding the problem altogether.  And I don't care about that misdirection to change the topic.  How a piece of music is produced is invariant to this discussion.  However it is created, a further conversion to 16/44.1 is not necessary.  Why do you guys have such a hard time with this concept?  For CD that conversion is mandatory since the format requires.  With digital distribution there is no such restriction so we can and are getting the pre-converted bits.


Amir
Retired Technology Insider

  • ajinfla
  • [*][*][*][*][*]
Audibility of "typical" Digital Filters in a Hi-Fi Playback
Reply #437
I am simply avoiding the problem altogether.




What problem? 16/44 is transparent, per award winning digital luminary BS/Meridian.
You certainly can't demonstrate any ability to audibly detect this imaginary "problem", not supervised. Never have, never will.

cheers,

AJ

Loudspeaker manufacturer

  • ajinfla
  • [*][*][*][*][*]
Audibility of "typical" Digital Filters in a Hi-Fi Playback
Reply #438
For CD that conversion is mandatory since the format requires.

Yep and audibly completely transparent. There is not a whit of evidence to suggest CDs aren't. So anything that cost more for the same 2 channels with zero audible benefit, is a $cam.

cheers,

AJ
Loudspeaker manufacturer

  • krabapple
  • [*][*][*][*][*]
Audibility of "typical" Digital Filters in a Hi-Fi Playback
Reply #439
I have a much easier solution that doesn't require new hardware from anyone: get the high sampling rate version of the content.  Problem solved.  You all can do the same if you choose.  Tell me again why you guys are arguing.
The false puffiness over the fear that SRCs will cause audible artifacts in 5 of 9 attempts under specific criteria, when far larger issues remain unresolved in the process as they relate to what is actually provided to the buying public?



For reasons we can only speculate about,  Stuart and the hi rez cheerleading squad (which maybe should be rebranded as the Redbook Fearmongering Corps)  would rather focus on eliminating the ant in the room.  It's odd because surely there is money in upgraded *mastering* and in *room treatment/correction*,  things which would address the real problems with modern home audio.
  • Last Edit: 21 November, 2014, 02:03:17 PM by krabapple

  • ajinfla
  • [*][*][*][*][*]
Audibility of "typical" Digital Filters in a Hi-Fi Playback
Reply #440
With digital distribution there is no such restriction so we can and are getting the pre-converted bits.

Ok amir, you keep (Red Herring) challenging other folks to take audio tests, including myself. I'll bite. I'll do a Hi-Rez comparison just like in the BS paper.
Where on the 2L digital distribution site can I find a restricted RPDF downsampled CD 16/44 version of the pre-converted bits Haydn track?
There is such a thing being distributed to end users as the BS paper has concocted, yes? This non-transparency distributed music isn't just an audiophile imagination bogeyman, right?
I'll get my transparent processing Meridian 518 player warmed up while awaiting your answer.

cheers,

AJ
Loudspeaker manufacturer

  • julf
  • [*][*][*][*]
Audibility of "typical" Digital Filters in a Hi-Fi Playback
Reply #441
It is a shame though that I can't get people to agree with the simple business principal that high resolution downloads are happening.


Oh, don't worry, we agree with that part. But marketing and sales has nothing to do with real benefits, something an ex-Microsoft person should be very intimate with.

  • mzil
  • [*][*][*][*][*]
Audibility of "typical" Digital Filters in a Hi-Fi Playback
Reply #442
This part of the BS paper always struck me as odd when testing something which quite possibly is related to very high frequencies, since 65 year-olds are typically expected to have high frequency loss:

"Eight listeners took part in the test, seven of whom were male. Most of the listeners were audio engineers, and their ages ranged from 25 to 65. All reported normal hearing, although this was not tested formally."

But I realize now the 65 year old, at the time of the testing, was most likely Bob Stuart himself [b. 1948]. Assuming he was also responsible for preparing the individual song segments he'd be in a unique position to be better prepared to listen for specific things like level mismatch, timing misalignment, sub-optimal dither artifacts including noise modulation, etc. and could have designed the song segment timings to make these specific issues particularly easy to ID.


   


  • Chu Gai
  • [*]
Audibility of "typical" Digital Filters in a Hi-Fi Playback
Reply #443
I have a much easier solution that doesn't require new hardware from anyone: get the high sampling rate version of the content.  Problem solved.  You all can do the same if you choose.  Tell me again why you guys are arguing.
The false puffiness over the fear that SRCs will cause audible artifacts in 5 of 9 attempts under specific criteria, when far larger issues remain unresolved in the process as they relate to what is actually provided to the buying public?



For reasons we can only speculate about,  Stuart and the hi rez cheerleading squad (which maybe should be rebranded as the Redbook Fearmongering Corps)  would rather focus on eliminating the ant in the room.  It's odd because surely there is money in upgraded *mastering* and in *room treatment/correction*,  things which would address the real problems with modern home audio.

It appears there is no need to speculate because the AES is working with the industry to bring hi-res to market. Consider the following two links:

http://www.aes.org/events/137/press/?ID=265
http://www.aes.org/events/137/specialevents/?ID=4222

Then take a look how Sony is marketing it.

http://discover.store.sony.com/High-Resolution-Audio/
  • Last Edit: 21 November, 2014, 03:04:32 PM by Chu Gai

  • ajinfla
  • [*][*][*][*][*]
Audibility of "typical" Digital Filters in a Hi-Fi Playback
Reply #444
It appears there is no need to speculate because the AES is working with the industry to bring hi-res to market. Consider the following two links:

http://www.aes.org/events/137/press/?ID=265
http://www.aes.org/events/137/specialevents/?ID=4222


Yes, it makes perfect sense that DEG, a group that is almost exclusively MCH would want to promote the benefits of "Hi-Rez" audio. Given that there is not a whit of evidence it can be heard in 2ch...and amirs other luminary, Dr Floyd Toole, found that audible issues (loudspeaker) discrimination went down, as more channels were added to the soundfield. So much so, that to make things easier, they test in mono.
I wonder if they use "Hi-Rez" mono??
Someone we know says they've been there....

cheers,

AJ
  • Last Edit: 21 November, 2014, 03:15:58 PM by ajinfla
Loudspeaker manufacturer

  • xnor
  • [*][*][*][*][*]
  • Developer
Audibility of "typical" Digital Filters in a Hi-Fi Playback
Reply #445
The answer is right in what I wrote but since it was misunderstood, I am going to explain the basics.

You don't need to explain the basics to me. I know signal processing basics, let alone science basics.
So let me correct where you get the basics wrong...


What you describe is not "ringing" but passband ripple.

No, I don't use technical terms ambiguously or even with the wrong meaning. I described filter ringing at over 21 kHz and not passband ripple.


Yes, it is sometimes called ringing but given the fact that I was talking about time domain (see the new highlight in red), you shouldn't have gone to frequency domain.

Given that fact it seems you don't know what you're talking about. Again, for the nth time, the filter shows ringing at over 21 kHz as visible in the impulse response (although frequency domain representation is equivalent anyway).


There is no disagreement there.  The paper makes it clear and I have said the same that the filters are near perfect in frequency domain:

For both FIR lters, the ripple depth over the passband was a maximum of 0.025 dB, and the stopband attenuation was 90 dB.

Note the correct terminology of "ripple" not ringing.  But yes, that is as ruler flat as we need it with just +-.025db variation and aliasing truncated by 90 db.  I wouldn't be able to hear that 0.025 db variation if it were in the range that I can hear.  And this as you say is in the range that I can't hear.  So again, no disagreement that that the frequency domain analysis doesn't indicate why this could be audible.

No, you are confused. I'm not talking about ripple but ringing. I never even mentioned ripple.
Also, 0.025 dB ripple does not mean +/- 0.025 dB. Ripple already specifies the max. variation.


What followed in my post was a hypothesis.  It is a hypothesis that is stated in the paper as I indicated.  As any hypothesis, it is being offered as a potential answer, not proof.  There is foundation in it but not one that I have personally investigated and hence the phrase I used: "I am starting to think..."

Yes, and I asked you how you could hear 21+ kHz ringing of a linear phase filter, that operates at 21+ kHz (except for the ripple, which seems negligible so I didn't even mention it in the first place).


What is that thinking?  It is what I described.  What is happening in time domain.  I created two filters in Matlab using the same 0.025 db passband ripple and 90 db out of band rejection.  The first one is very similar to what Stuart used in his study (not identical because I am not performing any optimization):

{img}

This is for the 48k sampling case.  We are filtering everything above 24 Khz and starting to do that at 23.5 Khz as the paper mentions.  This is not the frequency domain response which would be boring in its flatness as described above and in your post.  But rather the impulse response of the filter. 

Now we see "ringing."  The center is our signal that we fed to the filter (a sharp spike).  What we have now is pre- and post-ringing.  The impulse is replicated prior to its appearance (pre-ringing) and after its appearance.  The math dictates that it behave this way.  Indeed this is a computer simulation based on the mathematics of the signal processing.

Before I say more, let's look at the same filter, but this time we relax the constraints so that the filtering starts at 21 Khz instead of 23.5.  In other words, instead of giving the filter just 500 Hz to go from max to -90 db, we give it 3,000 Hz to do the same.  The picture changes dramatically:

{img}

Don't worry about the details of the graph.  The scales are different so this one is much more magnified.  Pay attention to red arrows here and in the previous graphs.  In this instantiation, ringing substantially subsides 0.2 milliseconds and 1.7 milliseconds.  So the difference is roughly 1.5 milliseconds.  This is the duration of our "distortion" if you will.

Applying the same analysis to the sharp cut-off filter that was tested in the listening test gives us 9.5-2.5 = 7 milliseconds.  In other words the span of ringing is now about 5 times wider in time than it was in the short cut off.

All of this was already pointed out in #362, but with more accurate numbers.

And no, this is no distortion. This is 21+ kHz ringing of a linear phase low pass filter.

The worst case (which Meridian claims to reflect real world filters - it doesn't seem to..) gives you 4.5ms for the total pre-ringing part. This ignores that the coefficients drop rapidly, for example in this worst case, to -20 dB within a small fraction of a ms.

Also, at the risk of repeating myself again, this ringing is at 21+ kHz. The filter operates at 21+ kHz and only on the energy that is left up there.



Is this audible?  I gave a reference in my post as to how it might be: pre-echo.  In psychoacoustics there is a concept of temporal masking.  If you hear a loud sound, what follows it very quickly may not be audible.  So if we look at our impulse itself, it creates a pretty nice shadow over what is to follow it.  But looking behind it, there is impulse (assuming silence for now) and backward temporal masking is quite limited.  This could, could, make the pre-ringing audible.

First of all, echoes are generally understood to be reflections that the listener can perceive individually as distinct events and therefore require a delay of several tens of ms to multiple seconds. That's why I dislike the term "pre-echo".
The filter above doesn't fit any of this, especially not the delay. Remember.. linear phase.

It may come as a surprise to you, but an ideal single tone rings infinitely at its specific frequency. Why am I telling you this?
Well, if you sum up tones up to a specific point, let's say 22.05 kHz, you will get an impulse response (and therefore FIR lowpass filter) that also rings infinitely. The ringing is that of the sharp cutoff at 22.05 kHz. Everything below that is passed perfectly without "distortion" or delay.
See #364.


Note that since this is in time domain, it has nothing to do with my ability to hear high frequencies.  All audio samples go through the above transformation, not just high frequency ones.  My truncated frequency response has no bearing on this situation unless the test tones are only above my hearing which is not the case in this musical segment.  I am pretty sure I can hear what music is playing .

Ringing at 21+ kHz of a filter... How often do I need to repeat this?
Take a look at #364. It shows a sine sweep from 20 to 23 kHz. As you subtract the filtered signal, nothing is left up to wherever the filter starts rolling off, which is above 21 kHz for the filters above.
This is a time domain waveform.

So how do you hear a difference?


The way you refer to pre-echo as "not being an echo" is totally nonsensical.  Of course it is not an echo.  It is *pre*-echo as it happens prior to the signal itself.

Exactly, it is not an echo in any sense hence I prefer the (imho more correct) term filter pre-ringing. Now if you weren't so confused about how filters operate...

The psychoacoustic component of the effect is that one hears only the echo preceding the transient, not the one following – because this latter is drowned out by the transient. Formally, forward temporal masking is much stronger than backwards temporal masking, hence one hears a pre-echo, but no post-echo.

It did it again! 

You mean you demonstrated your ignorance once again? The context of passages from the wiki you quoted talks specifically about lossy compression that makes use of frequency domain transformations. Simplified (just for you amir!) this is like chopping the signal into many frequency domain chunks.
Guess what this means? Many steep bandpass filters that cover the whole audible range from 20 Hz to 20 kHz. This is an entirely different issue than a lowpass filter operating at 21+ kHz.

Do you understand the difference?
  • Last Edit: 21 November, 2014, 03:57:31 PM by xnor
"I hear it when I see it."

Audibility of "typical" Digital Filters in a Hi-Fi Playback
Reply #446
And by downloading such bits I am free of worrying whether downgrading them is audible or not.
There is some logic in there. There is also some logic in capturing the microphone output as losslessly as possible, even if we can't hear it. DXD looks like a safe format for that, since most microphones don't have a bandwidth of more than 100 kHz. Higher-res downloads are more expensive and the claim is that the higher the bitrate, the better the sound quality. From older papers I got the impression that Bob Stuart seems to think that audible differences disappear above 24/96. Besides the satisfaction of "having the studio quality", what would be the point of buying higher formats ?

Some prices for an album in euro, from 2L.no , the provider of the samples of this paper:

stereo:
mp3      09.00
16/44.1   15.00
24/96   21.00
24/192      23.00
DSD64      28.00
DSD128    33.00
DXD          37.00
multichannel:
24/96        25.00
DSD64      33.00

(sorry, can't get tabs to work)

  • krabapple
  • [*][*][*][*][*]
Audibility of "typical" Digital Filters in a Hi-Fi Playback
Reply #447
It appears there is no need to speculate because the AES is working with the industry to bring hi-res to market. Consider the following two links:

http://www.aes.org/events/137/press/?ID=265
http://www.aes.org/events/137/specialevents/?ID=4222


'the full range of sound'...hmmmm.  Do they mean *sound* as in *what can be heard*?

Quote
Then take a look how Sony is marketing it.

http://discover.store.sony.com/High-Resolution-Audio/


This is actually clever of them:

Quote
Discover subtle details and artistic nuances in your favorite
music that you’ve never heard before. Feel the power
and presence of a live performance in your living room.
Or experience what it’s like to sit in on a live studio recording.
It’s all possible with the superior quality of High-Resolution Audio.
With quality greatly surpassing that of MP3 and CD, the difference is clear.


So, to cover their asses, they could claim to mean the slightly more revealing noise floor in the most quiet parts of a recording, or the very loudest parts of a recording with a dynamic range  >96dB  (probably a *live* recording) or the slight effects of mediocre antialias or antiimaging filtering


The only part where they back into bad habits is implying that the 'quality greatly surpasses' CD and that it translates to 'audible performance greatly surpassing CD's' and that the difference is 'clear' -- like, no training required, no special effort at 'discovery'.
  • Last Edit: 21 November, 2014, 05:39:40 PM by krabapple

  • krabapple
  • [*][*][*][*][*]
Audibility of "typical" Digital Filters in a Hi-Fi Playback
Reply #448
And by downloading such bits I am free of worrying whether downgrading them is audible or not.
There is some logic in there. There is also some logic in capturing the microphone output as losslessly as possible, even if we can't hear it. DXD looks like a safe format for that, since most microphones don't have a bandwidth of more than 100 kHz. Higher-res downloads are more expensive and the claim is that the higher the bitrate, the better the sound quality. From older papers I got the impression that Bob Stuart seems to think that audible differences disappear above 24/96. Besides the satisfaction of "having the studio quality", what would be the point of buying higher formats ?


IIRC Stuart thinks the 'problems' with Redbook disappear at ~55kHz, ~20 bits (not sure about the bitdepth, it's whatever Fielder claimed in his papers)  So either 88.2/24 or 96/24 , being common formats, were the  the logical choices.

I don't recall him ever advocating more than that, as being  either necessarily or useful.

It could however be useful for something like Plangent Processing, which relies on recorded ultrasonic bias tones to correct tape flutter.
  • Last Edit: 21 November, 2014, 05:37:45 PM by krabapple

Audibility of "typical" Digital Filters in a Hi-Fi Playback
Reply #449
On page 2 I find two rather conflicting statements and I'm not sure what conclusion the authors draw:
Quote
It has further been suggested [22, 23, 24] that listeners can discriminate timing differences of the order of 5 µs and below, which, if correct, would require a Nyquist frequency of 32 kHz or higher...

And below that:
Quote
However, it can not be assumed that the auditory system cannot extract time differences that are shorter than the periods between successive samples, even when convolved with a sinc function in D/A conversion (assuming a sufficient signal-to-noise ratio).

It is my understanding (references to JJ, David Griesinger e.a.) that "In Physics, the accuracy of timing is not determined by the bandwidth, but roughly by the product of the bandwidth and the signal to noise ratio".
Am I correct that the authors say that the time resolution argument based on sample rate and excluding SNR is not valid ?