[71 dB]

One of the significant problems is confusing visible and audible things. A perfectly band-limited impulse (sinc function) can easily be plotted visually to show "ringing", but that's visual representation of a mathematical function and we can select the axis so that what we see what graph tells about the function in question in a fruitful way. For example you can plot the sinc ( 𝒙 ) = sin ( 𝜋𝒙 ) / 𝜋𝒙 in the region -10 ≤ 𝒙 ≤ 10 and see the "ringing", but you could also plot it in the region -1000 ≤ 𝒙 ≤ 1000 in which case the function would look like an perfect impulse without any (visible) ringing. Now, how do we hear it?? More like the first case or the second case? The answer is more like the second case.

The "ringing" isn't even a problem to begin with, but a fundamental part of digital audio. It is the "weighting" function of how the original signal is reconstructed from the samples that represent the original signal spread in time. The ringing manifests itself only if we have signals that start of stop instantly, but those signals are illegal (not bandlimited). Sound of violin starts from zero and takes hundreds or thousands of samples to "get going". From the viewpoint of individual samples any sound meaningful for humans to listen to start and stop very slowly and what happens is the ringing of sample points one after other cancel each other away completely. Nothing "rings" with music signals and even if it did, it would be totally according to the band-limitation of digital audio.

Audiophools SEE visual depictions of digital audio. They don't understand how to interpret what they see and they don't undertand what we can hear and see are completely different things. Then comes snake oil sellers and spot the opportunity to make money, hence all the marketing BS we have to debunk here week after week, over and over again...

 

[Ghoostknight]

So the ringing just exists in the range where the filter is actually applied? (usually 20-22,05khz for 44,1khz samplerate)
i guess just like IIR filters alter phase in the corresponding range (and a bit above/under)

If i use resampling, and resampling adds the same kind of filter anyway in the same range as the dac would have (plus the dac adds now a second filter at the higher samplerate), is there even any difference?

i guess the only thing that could differ is that resampling actually utilizes a different kind of filter

i guess this makes the impulse response thing pretty insignificant, tho i still hear differences between different resampling "algorithms" and anti-image filters
maybe i should just go for what subjectively sounds best instead of trying to find the objectively best thing...

 

[bigshot]

Listen for what objectively sounds best. Numbers and charts on paper can give you an idea of where problems may lie, but a controlled listening test is the best way to determine whether it's audible.

 

[gregorio]

So the ringing just exists in the range where the filter is actually applied? (usually 20-22,05khz for 44,1khz samplerate)

That’s where the vast majority of the ringing is.

If i use resampling, and resampling adds the same kind of filter anyway in the same range as the dac would have (plus the dac adds now a second filter at the higher samplerate), is there even any difference?

No practical difference at all. Think about the logic. Let’s say that resampling/upsampling makes a JND (just noticeable difference), IE. The difference is too small for most to ever hear but those with good hearing/trained listening skills and good quality reproduction systems can hear it. If you were to resample/upsample/downsample say 20 times, then obviously that JND difference is now going to be blatantly obvious. And yet it isn’t, it’s still inaudible. As already mentioned, most recordings probably have at least a dozen re/over/up/down sampling operations performed during recording, mixing and mastering, many recordings probably have 2 or 3 dozen and some could have over a hundred.

i guess this makes the impulse response thing pretty insignificant

From a listening point of view it’s completely insignificant. From an academic and engineering point of view it’s not “pretty insignificant”, it’s potentially useful information and obviously it’s not insignificant from an audiophile marketing point of view because quite a few audiophiles have been convinced by BS marketing that the “impulse response thing” is an audible “problem” which is solved by some audiophile snake oil product!

tho i still hear differences between different resampling "algorithms" and anti-image filters

You can hear a difference between a single resampling/anti-image algorithm application but can’t notice when 20 have been applied?

If you really are hearing a difference (rather than just perceiving a difference due to a cognitive bias) then you must be using one of the deliberately sub-optimal filters. EG. One of the “slow”, early roll-off filters with a transition band starting around 10kHz or lower.

maybe i should just go for what subjectively sounds best instead of trying to find the objectively best thing...

The problem with subjectivity is that it’s affected by knowledge, experiences and preferences, which are constantly evolving and changing. Subjectivity can therefore change over years, months or sometimes even days/hours. It’s therefore best to go for as objectively/audibly high-fidelity (accurate reproduction) as you can and change it with something like EQ to match your subjective preferences because it doesn’t cost anything, except a bit of time, to change an EQ setting if/as your subjectivity changes.

G

 

[Ghoostknight]

No practical difference at all. Think about the logic. Let’s say that resampling/upsampling makes a JND (just noticeable difference), IE. The difference is too small for most to ever hear but those with good hearing/trained listening skills and good quality reproduction systems can hear it. If you were to resample/upsample/downsample say 20 times, then obviously that JND difference is now going to be blatantly obvious. And yet it isn’t, it’s still inaudible. As already mentioned, most recordings probably have at least a dozen re/over/up/down sampling operations performed during recording, mixing and mastering, many recordings probably have 2 or 3 dozen and some could have over a hundred.

its not like a specific resampling makes it necessarly "worse" (well it kinda does if you use for example with the spa resampling of pipewire quality 1 vs 15) but when i compare 2 resampling methods that are kinda on the same "high level performance" then its barely a different kind of "flavour" i would say... the question here for me is whats most true to the original (i guess no resampling is, but good resampling is pretty close, you wouldnt be able to say "hey i hear resampling!" for example)

where this really doesnt matter is in recording, why? because the sound engineer actually listened for it and he was happy with the result imo, he maybe even compensated to some of the added "flavour" of the resampling without even knowing
but in reproduction it kinda matters since you wanna stay true to the original and not add additional "flavouring"

but for me, overall, i have to say resampling in good quality can actually have (a atleast subjectevily) improvement over no resampling, atleast with 44,1khz files, so no resampling isnt necessarly the best either (what i kinda noticed was that the difference people usually refer to when speaking about high res vs cd quality kinda dissappears if you resample to say 192khz, if its not a different master) tho different clock performances for 44,1khz vs 48khz could be a reason here too

From a listening point of view it’s completely insignificant. From an academic and engineering point of view it’s not “pretty insignificant”, it’s potentially useful information and obviously it’s not insignificant from an audiophile marketing point of view because quite a few audiophiles have been convinced by BS marketing that the “impulse response thing” is an audible “problem” which is solved by some audiophile snake oil product!

well the question is why even measure something that is "completely" inaudible? i guess with very very worse filters it would be, and i dont think we can say for sure that audibility is for each the same, or even that studys made about this are even meaningful/tell the whole truth

If you really are hearing a difference (rather than just perceiving a difference due to a cognitive bias) then you must be using one of the deliberately sub-optimal filters. EG. One of the “slow”, early roll-off filters with a transition band starting around 10kHz or lower.

atleast with the ess9038q2m chip all filters seem to be ok regarding early slow roll... maybe one starts at 19khz but i think thats it but what worrys me more about the slow filters is that at the nyqist frequency some of them are not at their lowest filter point yet, which is objectively worse

i actually tried the brickwall filter again yesterday after reading the whole insignifact ringing thing and with 192khz resampling i actually prefer it... bass got more"punchy/direct" somehow

im 100% certain now that these different anti-image filters perform differently with or without resampling

The problem with subjectivity is that it’s affected by knowledge, experiences and preferences, which are constantly evolving and changing. Subjectivity can therefore change over years, months or sometimes even days/hours. It’s therefore best to go for as objectively/audibly high-fidelity (accurate reproduction) as you can and change it with something like EQ to match your subjective preferences because it doesn’t cost anything, except a bit of time, to change an EQ setting if/as your subjectivity changes.

yes i totally agree... i also suspect that many audiophiles (including me) sometimes prefer something that is actually "worse", this didnt just happen once to me

being a "purist" isnt always the best, specially regarding EQ... if i compare some of the differences i heared with lets say power supplys/dacs/other tweaks then this difference is maybe comparable to a 0.3 to 1db (broader band) eq change but these tweaks actually dont change the FR, i just say the heared differences are sometimes comparable like this, i sometimes imagine some tweaks make certain areas "smoother" and therefore we perceive them differently but this is just a guess... (even if this doesnt make sense since a 1khz tone is always a 1khz tone..)

overall you can change way more the character of the music with EQ, everyone that is denying this misses out heavly... maybe even more then using a switching power supply instead of linear x) and there will never be a room or a headphone that gives you perfect response, it might be "well enough" but thats it

but all that said, after trying kinda alot around with EQ i can still say FIR sounds better then IIR ... avoid IIR if you dont use it for crossovers
thats also maybe the whole culprit ... IIR can sound worse than without EQ in some ways and many if not most EQ`s actually utilize just IIR
the only easy to use FIR equalizers only exist as VST plugins i think

---

But yea.. i kinda just go with what my taste fits now with resampling/anti image filters, atleast with my dac the filters are all somewhat fine objectively speaking and also resampling "shouldnt matter" soo... i just go for what i like now

tho one huge red flag remains imo... why is ESS advertising different "sound signatures" with anti image filters? even dacs like the popular RME 1k€ one offer different filter settings
i mean i agree with it, they have different sound signatures but if all would be perfect why companys like RME dont just choose objectively the best? (probably brickwall right?)

 

[gregorio]

when i compare 2 resampling methods that are kinda on the same "high level performance" then its barely a different kind of "flavour" i would say...

Neither digital, analogue nor acoustic signals have any “flavour”. So whatever “flavour” you’re imagining is just that, an imagination! Now *maybe* your imagined “flavour” relates to an actual audio property (frequency or amplitude for example), in which case we can measure it, or maybe it is only an “imagination”, in which case it only exists in your head and we obviously can’t measure it. As we can measure differences between “resampling methods” but they’re outside of audibility, which option is left?

where this really doesnt matter is in recording, why? because the sound engineer actually listened for it and he was happy with the result imo, he maybe even compensated to some of the added "flavour" of the resampling without even knowing

How does “the sound engineer actually listen for” something that’s inaudible? And even if we could hear it, how could we compensate “without even knowing”? Don’t you think we’d notice instantiating say an EQ plugin and then adjusting it? And why do we commonly not see EQ or other adjustment/compensation plugins after resampling plugins?

but in reproduction it kinda matters since you wanna stay true to the original and not add additional "flavouring"

How do you think we mix or master recordings without reproducing them?

but for me, overall, i have to say resampling in good quality can actually have (a atleast subjectevily) improvement over no resampling, atleast with 44,1khz files

How have you compared resampling with no resampling?

what i kinda noticed was that the difference people usually refer to when speaking about high res vs cd quality kinda dissappears if you resample to say 192khz…

The difference “people usually refer to about hi res vs CD” is a difference only in their head/imagination or due to comparing different masters. Resampling obviously doesn’t create a different master, so you’re saying that resampling changes your imagination.

tho different clock performances for 44,1khz vs 48khz could be a reason here too

What different clock “performances”?

well the question is why even measure something that is "completely" inaudible?

To find out what it’s properties are.

i dont think we can say for sure that audibility is for each the same,

Science does not say audibility is the same for everyone, in fact science not only states it’s different for everyone but that it’s different for the same person because our thresholds of audibility change dynamically and deteriorate with age. However, we can obviously predict what is not audible because:

or even that studys made about this are even meaningful/tell the whole truth

If we measure a threshold of audibility, how is it not meaningful or telling the whole truth? You could argue that 1 test might not “tell the whole truth”, depending on sample size, but what about thousands (or millions) of tests over a period of 130 years or so?

i actually tried the brickwall filter again yesterday after reading the whole insignifact ringing thing and with 192khz resampling i actually prefer it... bass got more"punchy/direct" somehow

No, it doesn’t. We can easily measure/test the bass after a resampling operation and it’s literally identical! The place the signal changes is at and above the transition band of the filter, so the bass is literally the last place to look for differences! There’s only two other things that can change when over/up sampling an entire track: 1. The level (of the entire track) can be reduced by a small amount in some cases and 2. Your “imagination” can change!

im 100% certain now that these different anti-image filters perform differently with or without resampling

Are you 100% certain the filters themselves perform differently or that your imagination does?

if i compare some of the differences i heared with lets say power supplys/dacs/other tweaks then this difference is maybe comparable to a 0.3 to 1db (broader band) eq change but these tweaks actually dont change the FR

We can trivially measure differences of 0.3 to 1dB, so it would be trivially easy to measure these “tweaks” if they existed. And, if they don’t change the FR, what do they change?

i sometimes imagine some tweaks make certain areas "smoother" and therefore we perceive them differently but this is just a guess...

Exactly, you “perceive them differently” because you imagine/think it’s different, a classic example of perceptual bias. With such an obvious likelihood of a perceptual bias/error, how do know you are actually hearing a difference with these “tweaks” rather than just experiencing a perceptual error/bias? And, if you don’t know, then isn’t asserting you heard a difference dishonest?

tho one huge red flag remains imo... why is ESS advertising different "sound signatures" with anti image filters? even dacs like the popular RME 1k€ one offer different filter settings
i mean i agree with it, they have different sound signatures but if all would be perfect why companys like RME dont just choose objectively the best?

That “one huge red flag” is the same red flag we often mention in this subforum, MARKETING!

For the whole history of ADCs and DACs, they all did only offer one (objectively the best) anti-image/anti-alias filter. Then a few years ago, a chip manufacturer noticed some survey responses from audiophile manufacturers for switchable filters and as they had nothing else reasonable/audible to implement they added the facility to select different filters. The handful of other chip manufacturers didn’t want their chips to lack the facility compared to their competitor, so they added it too. You ONLY find this facility implemented in audiophile DACs though, there are no switchable filters in ADCs or in studio/pro-audio DACs. Lastly, “Sound Signature” is a vague (typically audiophile) term. The transition (and stop) band does vary with each different filter, so the FR is obviously different and therefore one *could* say the “sound signature” is different even though the difference is inaudible. However, it’s not uncommon for one of the filter options to have a transition band well within the hearing threshold. I believe one or two of the RME consumer models have a filter choice designed to (somewhat) emulate a NOS DAC, so the transition band would probably need to start around 2kHz (and obviously could be audible).

G

 

[bigshot]

Resampling from one transparent format to another isn't audible. If it is, there's something wrong.

 

[Ghoostknight]

Neither digital, analogue nor acoustic signals have any “flavour”. So whatever “flavour” you’re imagining is just that, an imagination! Now *maybe* your imagined “flavour” relates to an actual audio property (frequency or amplitude for example), in which case we can measure it, or maybe it is only an “imagination”, in which case it only exists in your head and we obviously can’t measure it. As we can measure differences between “resampling methods” but they’re outside of audibility, which option is left?

if resampling utilizes the same anti image filter as the dac would then resampling can change its "sound signature" too, just like the anti image filter of the dac

How does “the sound engineer actually listen for” something that’s inaudible? And even if we could hear it, how could we compensate “without even knowing”? Don’t you think we’d notice instantiating say an EQ plugin and then adjusting it? And why do we commonly not see EQ or other adjustment/compensation plugins after resampling plugins?

because they were still happy with it

How have you compared resampling with no resampling?

how would you test it? disable or enable it... in my case in camillaDSP and pipewire

What different clock “performances”?

jitter/phasenoise for example

To find out what it’s properties are.

kinda dumb to check something that is "completely" irrelevant tho

No, it doesn’t. We can easily measure/test the bass after a resampling operation and it’s literally identical! The place the signal changes is at and above the transition band of the filter, so the bass is literally the last place to look for differences! There’s only two other things that can change when over/up sampling an entire track: 1. The level (of the entire track) can be reduced by a small amount in some cases and 2. Your “imagination” can change!

well "imagination" stops for me when i can switch back and forth and the heared difference are identical...

Are you 100% certain the filters themselves perform differently or that your imagination does?

yes, the first one

no one said something about the fact that ESS writes in their own manual that "filters can change the sound signature", shouldnt they know better as probably the biggest chip manufacture for these kind of chips? shouldnt they avoid getting bad reputation for claiming audiophile BS ?

Exactly, you “perceive them differently” because you imagine/think it’s different, a classic example of perceptual bias. With such an obvious likelihood of a perceptual bias/error, how do know you are actually hearing a difference with these “tweaks” rather than just experiencing a perceptual error/bias? And, if you don’t know, then isn’t asserting you heard a difference dishonest?

this is BS.... if i change something and it stays for weeks exactly the same till i change something else than this isnt "perceptual bias", shouldnt bias of any kind change day over day? i dont think that your brain saves some sort of "setting" ....

I believe one or two of the RME consumer models have a filter choice designed to (somewhat) emulate a NOS DAC, so the transition band would probably need to start around 2kHz (and obviously could be audible).

never heared of such dacs/anti image filters... i guess this is real BS then..

 

[bigshot]

Changing "sound signature" is the same as saying, "changing frequency response, distortion or timing". Sound consists of frequency and amplitude arranged in time. If it is audibly altered, it can be measured. If there is no measurable difference, there isn't an audible one, because we can measure far beyond our ability to hear with our ears.

Bias doesn't necessarily change over time. It can last for weeks and weeks. It can even continue after it's proven that a difference doesn't exist. Bias isn't random error, it's a skew in a particular direction. It can be very persistent.

 

[Ghoostknight]

Changing "sound signature" is the same as saying, "changing frequency response, distortion or timing". Sound consists of frequency and amplitude arranged in time. If it is audibly altered, it can be measured. If there is no measurable difference, there isn't an audible one, because we can measure far beyond our ability to hear with our ears.

well if there is a audible difference AND it can be measured then we should question the audiblity

 

[bigshot]

There are thresholds of audibility. The measurements have to reach a certain level to be audible.

 

[castleofargh]

well if there is a audible difference AND it can be measured then we should question the audiblity

We do. For example, people claim audibility, and we question it ^_^.
Asking an audiophile how confident he is, equals to asking him how much self-esteem he has and how much he agrees with the rational he created in his own head. In science, confidence relates to how solid and conclusive the demonstration is. A sighted experience is nowhere near as conclusive as audiophiles and right now, you, want to think it is. Putting that aside isn't the answer.

You want to test the audibility of oversampling, start by trying to reduce the unknown variables. Take a file, oversample it with some method, downsample it back and do an ABX in foobar against the original at the same resolution. Everything else will be the same, and you have twice the resampling done to the file. If you get a clear difference, it might be worth it to check if there are extra options that aren't actually resampling and disable them. If nothing like that exists, then it might be a great idea to never use that resampling tool again.
But if you get about 50/50, well then you'll have to accept that other poorly controlled experiences probably had other variables involved(including perhaps you looking at the settings).
Oversampling files before sending them to the DAC is another problem. A good deal of the up/over sampling going on in the DAC has to do with reclocking and general anti jitter solutions. You might improve an old NOS DAC because IMO, NOS is stupid, and you're doing part of the necessary work for it. But you might lower the performances of a modern DAC by reducing the amount of oversampling it was going to use for a specific purpose in an already optimized system for that particular design. I'm guessing that could end up audible in some fairly bad scenarios, but the cause, while being a consequence of oversampling, is also not really oversampling but what oversampling deprived the DAC of.
Or it could maybe become a matter of already weak buffer running out of space faster because you have triple or quadrupled the streaming rate, or.... Complex systems are complex, and I wouldn't claim to know how to model all of them entirely(spoiler, I clearly can't).

If you want to test filters, you should try to check what they really do with some interconnect and REW, audacity or whatever RTA(real time analyzer) you can get for free. And if you really can hear them(not just deciding that you can after a casual sighted experience).
2 of my old DACs give me this as impulse response(worst possible looking impulses by using 44.1kHz and the 2 most different looking ones I ever owned):

"hermagerdz ringgrinkggzz!!!!!"
I guess I should have started with showing you that when talking about impulses and concepts of fidelity. Those who know anything about filters know what's going on here, as it's a textbook example for filter types.

And of course the massive FR difference:

My sighted attempt to tell those DACs apart resulted in me being sure of only one thing, the yellow guy had less bass... I swear I didn't notice anything else repeatedly.
In a single blind test(both with an almost perfectly identical 1V output, more by chance than thanks to a good test setup), I got random guess stats. Audible ringing, pre ringing, the bass I was so sure I was noticing being louder? Nope, I didn't consistently notice any of that in any of the tracks I picked for my tests.
And I previously trained myself with some extreme EQ scenarios using various filters, which did give me audible ringing, so I'd have some vague concept of what it feels like. But when I used them, of course it was in the low or midrange with high amplitude and the most extreme Q values my EQ could give me(I remember some funny effect on a drum sample with like a small soft muffled hit before the main louder hit, my first time going "oh! So that's pre ringing"). But trying that stuff at 20kHz with EQ or with those DACs, did nothing to me unless I started to roll off the treble pretty hard inside the audible range due to the filter setting. I'm guessing the other way around is also possible. Get so much aliasing that it's clearly audible too, but I never properly tested for that as I don't care. I don't purchase the kind of DAC that would think it's a good idea to massively fail band limiting on purpose. To think there was a time when some DACs didn't have a filter at all... Genius.

Admittedly, even after training, I remained not very good at it. It's not like background noise, clipping or some specific lossy artifacts that once you notice them, you start hating them and feeling like you can never unhear them. With ringing, even pre-ringing which isn't the most natural thing there is, I honestly don't mind. Kind of like I have a thing for instruments played backward, it sounds weird, but to me it's a nice weird. It doesn't jump at my throat like some really hard clipping would. All that to say, maybe it bothers some people more and even a little pre ringing ruins their experience(if they can actually perceive it)? IDK, I only tried me.

 

[Davesrose]

Now that I see that graph, I know my ears are sensitive to that extra .05dB at 22Hz and 5kHz ^-^

 

[Ghoostknight]

Asking an audiophile how confident he is, equals to asking him how much self-esteem he has and how much he agrees with the rational he created in his own head. In science, confidence relates to how solid and conclusive the demonstration is. A sighted experience is nowhere near as conclusive as audiophiles and right now, you, want to think it is.

imo hearing "differences" on a somewhat resolving system is way easier then to say what actually sounds superior (and is objectively superior), thats one of the reason why i myself are careful with reviews and other opinions, subjective opinions kinda suck and i understand why some prefer to lean completely to the objective side

i dont know... subjective expierences should always be taken with a grain of salt, they are subjective after all but if i hear consistent changes with switching back and forth (specially for days/weeks etc) im atleast pretty sure "that there is a difference" whats in the end actually better is the 100$ question in most cases

this test is really interesting, i wanna see more such tests! https://archimago.blogspot.com/2015/07/the-linear-vs-minimum-phase-upsampling.html
i found it while searching about SOX minimal phase vs linear phase "feature" because i was kinda surprised of the phaseshift (https://src.infinitewave.ca/?Top=Pipewire_Q14&Bot=SoX14_VHQ_MP&Spec=0144) as you see the phaseshift starts at around 2khz and is at -180° at 14khz, i wonder if the minimum phase filters of any dac have that huge of a phase shift?

it isnt really conclusive after all (since this is highly subjective...) but there are 2-3 somewhat conclusive results in the test made which is kinda interesting and it shows atleast a slight tendency in some areas

but for me this also kinda shows that with subjective stuff like this you will probably never get conclusive results for any study made about this

 

[bigshot]

Hearing differences is easier when you are comparing without applying controls.