Audibility of digital reconstruction filters

[dc2bluelight]

I want to concentrate on the audibility of different filters built by experienced people. That's simply because it's quite possible whatever combination of parameters I didn't like the sound of in iZotope SRC (or what I *did* like the sound of) is something no pro would ever do. I would expect less obvious variation in the sound of filters designed by pros.

 

That being said, I was floored by the degree of audible variation among the filter options provided in Miska's HQPlayer software. I would urge anyone curious about the topic of this thread to try it.

 

Agreed with the above. I would assume that a programmer or manufacturer providing a control would make darn sure it did something audible, or it probably wouldn't be there. If someone did an impulse measurement on these things, that would be really interesting.

[Superdad]

If someone did an impulse measurement on these things, that would be really interesting.

 

You mean like the sort you can look at at SRC Comparisons ?

 

 

I cited some references that showed a few things that have been proven to not be as audible as we thought.

 

What did you cite? The boogyman of "bias?"

 

I know from listening to a lot a DACs--and from improving on generic filters with s/w--that filter design can account for considerable difference between DACs. It's by no means the only important thing, as there are a hundred other design and implementation variables that are factors in the final sound.

 

So if folks want to discuss filters, let's do that. If it just turns into an argument about audibility then I'll be happy to tune out. To me, filters are as easy to hear as the difference between MP3 and WAV, but whatever...

[Jud]

If someone did an impulse measurement on these things, that would be really interesting.

 

I have an idea for experiments along these lines, but I would like to run it by some people privately first.

[Paul R]

(*sigh*)

 

You are welcome to your own opinions. Fact's won't change all that quickly.

 

I would be interested in your opinion as in "Yes, I tried modifying filter settings and I could hear a difference. I do not think this is caused directly by ringing in the filter, I think it is ... "

 

 

 

...except that's not all you're changing...

 

 

 

Huh. OK, hit the nerve there, didn't I?

 

I never said, "people are not hearing a difference, or that it is not possible for people to hear a difference. Or more perniciously, that people are imagining such differences due to expectation bias". Perhaps a re-read? All I'm saying is we need to know for certain they are hearing a difference. Is that so unreasonable? I also said, if they do hear a difference, we can hypothesize as to why, then go find out for sure. Is that also so unreasonable?

 

I cited some references that showed a few things that have been proven to not be as audible as we thought. Is that the problem?

 

Look, kids, you hear what you hear. I do get a bit tired of betting shot at just for trying to apply a bit of science and reason to find out why.

 

 

 

I certainly agree with that. But in the quest for audio nirvana, wouldn't it benefit everyone to know what sounds better and why, that tech can be integrated into more stuff?

 

But if it makes everyone happy, I will gladly accept the "evil scientist" label.

 

...and down I go...down my own personal rabbit hole.

[dc2bluelight]

What did you cite? The boogyman of "bias?"

Ha! Funny!

 

I could be mistaken, but I don't think either reference even used the word "boogyman".

[Paul R]

It depends a lot on what you are trying to test. Do people hear the difference? Yes. Does that mean the difference was caused by purely physical factors? Probably not. But in either case, people are hearing a difference.

 

The idea of exploring the audibility of filters is interesting. But how one goes about it is also important.

 

I think a more precise way to say it that "people claim to perceive a difference". Can they really, well that needs more than just asserting that people can actually hear a difference.

[esldude]

Not sure exactly how Miska meant his blurring comment in regards to film. I would think it one of two ways. Aliasing like the old Western movies where the wagon wheels looked to be turning the spokes backwards. Which is....well....aliasing by inadequate filtering. Or if the frames had high frequency info being spread out in time (read slightly into the UV region). If the latter, it is less accurate, but also not seen by us.

 

I think it is measurable, just not audible.

 

Here is a contrived example. I took the residual ringing of a 71 tap brickwall filter used upon impulses. That would be about 1.6 milliseconds in length of ringing or correspond to shaped transition band ringing modulated at 621 hz. Does this produce noise or artifacts down frequency due to the shape and length of the ringing?

 

Yes it does, but at such a low level I find it hard to think it audible. Remember this would be maximum ringing by a single sample impulse strung together and repeated 621 times per second. I can't see that happening in real music recordings. Also remember lower the abrupt level changes causing ringing to a lower level and these residuals are lower as well.

 

Here is a pic of the wave I created. Upper trace is the normal view, and lower is the waveform(dB) view so you can see I copy and pasted all of and just the ringing from the impulse to string together. This also was amplified 15 db to make it easier to see.

 

 

Next is an FFT of this created file of the ringing residual of a single sample impulse. The length of the filter corresponded to 621 hz and what you see is 621 hz and every harmonic of 621 hz above that. It only exceed minus 120 db around 18 khz and above. These spikes in the FFT were very narrow. So even a lower bin number of FFT didn't raise the levels. There is minus infinity between those spikes until you get to 18 khz. Above 18 khz what is between the spikes in the FFT is below - 200 db.

 

 

So there is something there, but I have a hard time thinking people are hearing that. And this is an extremely unlikely result for any impulse or transient induced results occurring in normal musical recordings. Not likely such a thing is possible with the anti-alias filter in an ADC.

 

I took the above contrived file, listened through some sensitive headphones with everything maxed on gain. I was starting to hear a bit of white noise of the equipment just barely. Were I to play a normal music file at this level I would endanger my hearing with likely 120 db plus output. The contrived file was inaudible. I boosted it by 10 db and thought I was starting to hear it. I boosted it another 10 db and was hearing it. I created a 621 hz sine wave, and set it at -75 db. It was still much louder than this contrived residual. I put one file in the left and the other in the right of a stereo file. I lowered the sine wave to -91 db before it got close to the volume of the residual though it was still louder. The frequencies did indeed sound like a match with the impulse residual sounding a bit rougher like you would expect with the harmonics. So after 20 db of boost this file was maybe around -100 db or lower. You are not hearing this while playing music. And I can't see how this amount of unintended result would occur even this high in level with an actual music recording.

[Miska]

After all, the only difference between filters is post, pre, or no ringing, at least as far as your experiment and derived theory go.

 

Of course there is much more than that... There is the filter pass/stop-band points, roll off slope shape, pass band characteristics (ripple, etc), stop band characteristics, different characteristics of rounding errors due to cascade multi-pass or single-pass design. Type of arithmetic and actual arithmetic construction vs the used computation hardware properties, etc, etc...

[esldude]

Science would say first, "Is there a difference?" To answer that, sadly, the experiment must eliminate biases of all kinds. Once established that there is a real and repeatably detectable difference, a hypothesis or three might be formed, then a proof sought.

 

Ah. If only that were true and it was that simple.

 

 

"Good Science" must by definition take into account the errors in the entire experimental system. Unfortunately, some of those errors include bias in observation, but there are many others as well. If as many errors as possible in the system are to be accounted for, then influence of biases must be included.

 

We could present hypotheses all day, but that would be less than satisfying. A hypotheses is only a beginning, and the scientist must go on to experimentally prove it. Much of this work has been done, though not published in main-stream press.

 

I agree with you dc2bluelight. However, since the issue is contentious and proper blind testing is not at all convenient, the thread was sort of in between the proper scientific path. People say there is a difference in filters that is audible. And that ringing or pre-echo (yes not really accurate descriptions especially for band limited squarewaves) is often given as the reason. So is the pre-echo or ringing audible?

 

So for the sake of internet discussion I am of course skipping over the proper step of "do people really hear this". Going instead to the middle step of what is the result of said pre-echo and ringing? What artifacts are produced? Do they really align with causing the differences claimed audible by some. Thus far it appears the artifacts of digital filters mostly are above any frequency we can hear. Some other artifacts are possible, most highly improbable to actually occur, and if they did are so low in level they could not matter. As in the original post perhaps some ringing could interact with downstream equipment deficiencies to be audible though again that looks likely so low in level as to be a non-issue with most equipment.

 

I would say there is no chance this thread will be a definitive answer to change people's minds in either direction. But perhaps for those trying to understand it (like me) the involved discussion will provide food for thought, maybe actual understanding, and a chance to consider what is really happening vs what is being claimed.

[esldude]

It depends a lot on what you are trying to test. Do people hear the difference? Yes. Does that mean the difference was caused by purely physical factors? Probably not. But in either case, people are hearing a difference.

 

The idea of exploring the audibility of filters is interesting. But how one goes about it is also important.

 

Well Paul, if your version of people hear differences is yes, and probably not caused by purely physical factors the obvious question is are there any physical factors? How much of non-physical factors are involved if some are physical and some are not. The only part of changing filter parameters that can be applied across different people are the physical factors. One cannot predict what a filter change would sound like if the difference heard is part physical and part non-physical.

 

BTW, do they have a local society of sophistry in your town, and are you the president?

[Miska]

Not sure exactly how Miska meant his blurring comment in regards to film. I would think it one of two ways. Aliasing like the old Western movies where the wagon wheels looked to be turning the spokes backwards. Which is....well....aliasing by inadequate filtering.

 

No, I meant motion blur. Each frame containing also weighted sum of previous frames. Or in this case certain number of future frames too. (make a google image search on "motion blur")

 

Linear-phase filtered output is weighted sum of current and N/2-1 previous and future samples. That's why step response shows influence of the step already N/2-1 samples before and also N/2-1 samples after the step.

[Jud]

Dennis, the differences between filters in HQPlayer, in non-blinded listening, seemed far more audibly obvious to me than I had assumed they would be. (I can think of blinded tests to verify this, but don't know if Miska would have the time or inclination to play along.) That's why I urged people to try it, because I think it might help reset skeptical expectations to something more like neutrality - "Hmm, think I hear something, do I really, and if so what could cause it?"

[Miska]

So is the pre-echo or ringing audible?

 

Ask any designer of a lossy codec (Vorbis, MP3, AAC) and they can spend days talking about this subject...

 

It is one reason AAC uses shorter window than MP3...

[Miska]

Next is an FFT of this created file of the ringing residual of a single sample impulse.

 

You are now chasing wrong thing. You need to look at time-frequency analysis at 50 µs window lengths and see if the transient's time envelope stays unmodified while at the same time keeping the frequency envelope within Nyquist boundaries.

 

And remember, hearing has better time-frequency analysis performance than you can get with FFT...

[esldude]

Ask any designer of a lossy codec (Vorbis, MP3, AAC) and they can spend days talking about this subject...

 

It is one reason AAC uses shorter window than MP3...

 

Well of course pre-echo can be audible. I was thinking in the context of reconstruction filters. MP3 has pre-echo at sub-20 khz and levels which can be audible. Not the same as the reconstruction filter of wav files. I have already said downsampled music to 8khz sample rates and the ringing in the vicinity of the 4 khz cutoff could be heard.

[Superdad]

Going instead to the middle step of what is the result of said pre-echo and ringing? What artifacts are produced? Do they really align with causing the differences claimed audible by some. Thus far it appears the artifacts of digital filters mostly are above any frequency we can hear. Some other artifacts are possible, most highly improbable to actually occur, and if they did are so low in level they could not matter.

 

Hi Dennis: You (and others) might find some of PeterSt's dirac impulse captures quite interesting: Phasure NOS1 USB Special Measurements

It is a long post, but scroll way down to about the bottom 15 graphs. It is not all about just his "arc prediction" filter. I found it kind of eye opening to see some of these differences resolved.

 

I can't tell you why I can hear the changes made by small changes in filter settings, but having tuned a few both alone and with others, I can assure you these things are important. Also, given the nature of what I hear (related to transient attack and also bass propagation) I am not sure if headphone listening is a good way to hear these changes. I admit to not being a big headphone listener, so maybe that is why I have a hard time judging the presentation full-sized instruments on them.

[4est]

Very much agreed. One should read his suggestions as a start.

That being said, I was floored by the degree of audible variation among the filter options provided in Miska's HQPlayer software. I would urge anyone curious about the topic of this thread to try it.

[dc2bluelight]

It depends a lot on what you are trying to test. Do people hear the difference? Yes. Does that mean the difference was caused by purely physical factors? Probably not. But in either case, people are hearing a difference.

 

The idea of exploring the audibility of filters is interesting. But how one goes about it is also important.

 

We should probably clarify something here. When you say, "Does that mean the difference was caused by purely physical factors? Probably not.", are you saying that the difference may be caused by something that is not physical?

 

What would "not purely physical" include? Are you referring to something metaphysical, spiritual, etc.? As a reference, sound, the hearing mechanism, transducers, and associated electrical energy is all physical. Are you going outside of that domain?

 

I'm not trying to trap you, and no judgements, but it may change the tack here just a bit to know...

[InfernoSTi]

This thread is really interesting to read when folks are trying to understand the issue at hand, even (especially?) when they disagree. Not so much when there is dismissive commenting being made, though…

 

Please keep discussing it!

 

John

[Fokus]

 

And remember, hearing has better time-frequency analysis performance than you can get with FFT...

 

only when comparing apples and oranges. There is a fundamental information-theoretical difference between an FFT and asking people in tests to discern between differently timed peeps or blibs. Jumping to the conclusion that the ear breaks the Gabor limit is just sensational writing. Gabor didn't apply in the first place.

 

As for the 50us thingy ... You aware of the time constants of the cochlear filters?

[dc2bluelight]

And remember, hearing has better time-frequency analysis performance than you can get with FFT...

 

It depends on what you look at. Since an FFT's frequency resolution depends on time (actually as do all types of spectrum analyzers), and hearing much less so, hearing can win a some things loose at others. Hearing can pick out the frequency of short duration notes mixed into complex music, FFT not so much. Yet FFT isn't affected by masking, so it can display a low level tone close in frequency to a high level tone given sufficient time. FFT can also be configured for wide-bandwidth flat response analysis, hearing has a wicked response curve that gets worse as levels drop.

 

I would not be correct to advance that hearing is simply better than FFT, therefore there are things that can't be analyzed. FFT can beat hearing in many ways. The uncertainty principle applies to FFT, but apparently not to hearing. In the end, if the differences are understood, the advantages of each can be exploited.

[Miska]

Yet FFT isn't affected by masking, so it can display a low level tone close in frequency to a high level tone given sufficient time. FFT can also be configured for wide-bandwidth flat response analysis, hearing has a wicked response curve that gets worse as levels drop.

 

I would not be correct to advance that hearing is simply better than FFT, therefore there are things that can't be analyzed. FFT can beat hearing in many ways. The uncertainty principle applies to FFT, but apparently not to hearing. In the end, if the differences are understood, the advantages of each can be exploited.

 

Of course, but that's not about time-frequency resolution I was talking about.

 

WVD is better than FFT on that area. And Gabor (transform) is good way to improve STFT. This is a pretty good book on the topic:

http://www.amazon.com/Joint-Time-Frequency-Analysis-Method-Application/dp/0132543842

[Miska]

only when comparing apples and oranges. There is a fundamental information-theoretical difference between an FFT and asking people in tests to discern between differently timed peeps or blibs. Jumping to the conclusion that the ear breaks the Gabor limit is just sensational writing. Gabor didn't apply in the first place.

 

I was referring to:

Human hearing beats the Fourier uncertainty principle

 

We are designing systems for people listening music, not for spectrum analyzers. Hearing is not FFT analyzer, it is more like a wavelet filter bank optimized for analyzing properties of the initial transient impact, first milliseconds of the waveform. So don't design you algorithms for FFT. Although it should of course look fine on FFT too.

 

I've been listening and analyzing so many transients like active sonar pings, klangs and clunks over my career (and teaching others on the topic), I know pretty well how hearing manages vs analyzers.

 

As for the 50us thingy ... You aware of the time constants of the cochlear filters?

 

If we assume 20 kHz upper hearing limit and want to be certain about preserved temporal resolution, 1/20e3 is the time constant we should be looking at.

 

Technologically, there's no reason why not aim for 20 µs temporal resolution to have safety margin. Completely achievable, no reason not to do it.

[Don Hills]

You 're at CA, not HA (hydrogenaudio), and are talkin' Greek to THIS crowd. So go to that place for your query ... and don't come back

 

Quite right. Everyone knows that different filters sound different. It's night and day.

.

.

.

 

I agree with both posters would like to third this sentiment. ...

 

In the interests of accuracy, I wish to point out that that I don't actually subscribe to the view that "different filters(*) sound different". Can anyone suggest a good smilie for "tongue in cheek" or mild sarcasm?

 

My own view is that:

(a) any "ringing" occurs at too high a frequency to hear,

(b) the "smearing" referred to, although it can occur over a time long enough to perceive, only "smears" frequencies too high to hear, and

© occurs at too low a level to be audible (masked by the content).

 

Edit:

(*) As used in A/D and D/A converters. Further, I am assuming that the filters are designed to perform their intended function and not deliberately altered to colour the sound. Not that I am implying that any filter designer would deliberately do this...

[Paul R]

Well Dennis - if you contend that the only part of hearing you can measure or predict is the physical part, how do you ever intend to get a real answer to your question? Or is the intent just to aggravate people with pseudo-science?

 

Ignoring data about an issue can very easily lead one to incorrect conclusions. I think it has already been pointed out that people can hear changes in ringing settings on filters, and that there are many more possible factors that can or do make audible changes. Yet you seem to ignore this and go on assuming you have proven there are no measurable audible differences. Not even asking how to duplicate the changes other people report.

 

You definitely must be a member in good standing of your local JumpToConclusionsWithoutProperDataOrUnderstanding Club. Perhaps a founding member. I wish I could better understand what your agenda is - because it sure looks to me you started this thread in another attempt to "prove" something you believe and convert the unwashed asses.

 

Tell you what though - I am out of it. Every person has the right to go to hell in his own way. Enjoy the ride.

 

Paul

 

 

Well Paul, if your version of people hear differences is yes, and probably not caused by purely physical factors the obvious question is are there any physical factors? How much of non-physical factors are involved if some are physical and some are not. The only part of changing filter parameters that can be applied across different people are the physical factors. One cannot predict what a filter change would sound like if the difference heard is part physical and part non-physical.

 

BTW, do they have a local society of sophistry in your town, and are you the president?