Head-Fi.org › Forums › Equipment Forums › Sound Science › Can someone explain audio resolution to me?
New Posts  All Forums:Forum Nav:

Can someone explain audio resolution to me? - Page 2

post #16 of 77

you don't need to equal or beat 24 bit - 24 bits theoretical noise floor already way beyond any recording technology, playback electronics ability, recording studio room or microphone noise floor when 0dbfs is set at 120 dB SPL

 

very good electronics, audio DAC, amplifier S/N seldom exceed 120 dB ~20 bit dynamic range (audio bandwidth, no weighting)

 

home listening rooms will have ~20 dB SPL noise floor so a system peaking at 120 dB SPL would only need ~100 dB dynamic range with proper system gain structure

 

 

so the 1st assumption about 24 bit audio for home playback is that you're wasting at least the bottom 4 bits


Edited by jcx - 8/17/13 at 11:35pm
post #17 of 77
Quote:
Originally Posted by UltMusicSnob View Post

There ARE real differences in playback sound between music CD audio and 192/24. I can hear those differences, so I upsample my music CD's to full 192/24. Yeah, it's a lot of space on the HD, and it takes time---that's my problem, no one else's.

 

The upsampling algorithm might alter the signal in a non-obvious but still noticeable way during critical listening. However, there is no new data in an upsampled 44.1/16 file, the algorithm can't conjure more data between the sample points. Thus what you're listening is either resampling, or your DACs different performance at different sample rates. However may it be, in the end the credit doesn't go for the 192/24 as a resolution. I've seen some DAC manufacturers quote different performance figures at different sample rates, usually worse for the "HD" sample rates, so maybe your DAC is interfering with the testing? The Benchmark DAC2 seems to promise same performance for all sample rates, if you're dead serious about proving HD formats you might want to look into it!

 

PS: although I'm skeptical about the significance of big sample rates, have you tried some of these super-high samplerate upsampling DACs? They do the conversion on the fly, you might (subjectively speaking) enjoy one a lot ;)

post #18 of 77

^ It is true that many chips do not perform as well (usually just small differences though) at very high sampling rates. That's one reason why the Benchmark DACs resamples internally to a fixed 110 kHz iirc.

post #19 of 77

Thanks for confirming.

But yes OP, in all simplicity sample rate divided by two is the maximum frequency that can be reconstructed from a sample file, it doesn't matter in any other way. 22050Hz is way over what any normal person can hear. If someone says they can hear above that, they're very likely cranking volume up on their gear, and their DAC, amplifier or headphones is having some serious distortion issues, leaking second, third and fourth order harmonics into the audible range (distortion tends to rise in the extremes of the audio band). Now THAT can be audible. There is no extra benefit from a higher samplerate or bit-depth, no "smoother" stairstep, because the sigma-delta DA converter reconstructs the wave perfectly thanks to filtering. (As suggested earlier, check out the output of an analog oscilloscope if you don't believe us)

I'm not trained on maths, but I could imagine higher samplerates being beneficial during mixing/mastering to reduce aliasing errors between mixes thanks to different sorts of digital effects and VST plugin that might not always have the most accurate rounding or something. However, I'm quite sure this too is trivial for the hearing, nobody has absolute pitch like that.

The only case where I know for a fact the bit-depth matters, is digital volume control. If you have a 16-bit DAC, and you attenuate the volume by more than 90% in both OS and player, you start seeing stair steps from the quantization error. That would require you to crank your amplifier a LOT, so it would be nonsensical to use your system like that and claim the ludicrous benefits of 24-bit (because even then, the noise floor is very close to being bigger than the quantization error, and in most cases it is).

 

However I would like to point out, and that a 24-bit DAC is going to convert 16-bit streams to 24-bit anyway on the OS level, and you most definitely don't need 24-bit HD tracks to enjoy this benefit.

 

post #20 of 77
Quote:
Originally Posted by Cat Face View Post

 

The upsampling algorithm might alter the signal in a non-obvious but still noticeable way during critical listening. However, there is no new data in an upsampled 44.1/16 file, the algorithm can't conjure more data between the sample points. Thus what you're listening is either resampling, or your DACs different performance at different sample rates. However may it be, in the end the credit doesn't go for the 192/24 as a resolution. I've seen some DAC manufacturers quote different performance figures at different sample rates, usually worse for the "HD" sample rates, so maybe your DAC is interfering with the testing? The Benchmark DAC2 seems to promise same performance for all sample rates, if you're dead serious about proving HD formats you might want to look into it!

 

PS: although I'm skeptical about the significance of big sample rates, have you tried some of these super-high sample rate upsampling DACs? They do the conversion on the fly, you might (subjectively speaking) enjoy one a lot ;)


Thank you for the tips! Of course it is true that no new data is created by upsampling--I assume that going in. And if there *were* any data in upper frequency ranges, it would not be heard directly. It must be the case, if audible differences at playback output exist [ they do, I have a long record of ABX positive results comparing 192/24 to 44.1/16 under multiple conditions on multiple platforms ], then processing **somewhere** between the original 44.1/16 data on the published disk, and the analog sound produced, must be responsible.

 

Right now I can get the results I enjoy essentially for free, using technology I already own: a music CD, SoundForge 10 for resampling, an audio interface that plays 192 (RME Babyface), and lots of hard drive space to spare.

 

It's conceivable that going beyond 192 might add something (what is super-high sample rate? --still in PCM, or is this SACD DSD?), but I'm not sure how that would be accomplished. As it is, I'm not at all sure why I experience a subjective improvement--I suspect temporal resolution effects during D/A, but of course I have no way to prove that directly.

post #21 of 77
Quote:
Originally Posted by UltMusicSnob View Post

what is super-high sample rate?

since DSD is not directly comparable to PCM streaming as far as sample rate goes, I was mostly referring to the likes of Cambridge Audio DACMagic Plus, which is an objectively excellent DAC for the price, that many people enjoy subjectively too. It has a fixed internal upsampling frequency of 24-bit/384kHz, and you can adjust the filtering used. For the technical tinkerer it might offer a small novelty factor. It's only disappointing aspect is the internal headphone amplifier, but as a DAC it seems quite sweet! Here are some measurements if one wants to indulge themselves in some objectivism: http://kenrockwell.com/audio/cambridge/dacmagic-plus.htm#meas


Sorry for derailing the original topic. Although I'm suggesting UltMusicSnob an oversampling DAC because I happened to remember one, I can't vouch for the importance or audibility of any of the methods described. I've personally never heard an improvement in an ABX test with different sample rates, and even if it was perceptible (which so far nobody has proved in controlled environments), I don't believe it's significant enough to make me enjoy regular 16/44.1 music any less. I hope OP has learned something of the two opposing opinions of people and the basic theory behind why and what people hear/believe.

post #22 of 77
I think he's hearing some sort of artifacting somewhere in the conversion process that he likes. A properly functioning upsampling DAC probably wouldn't do that.
post #23 of 77
Holy cow, 384 kHz...I remember that number from somewhere, but not from a PCM rate, I don't think.
The science on Nyquist relationship of sample rate to freq is solid, but it's a mistake, and an oversimplification by far, to think that A/D and D/A engineering, the science of acoustics, and the science of human hearing and mental processing is all complete. Far from it. The unavoidable D/A process alone, at any sample rate, involves tradeoffs among multiple parameters. If someone figures out how to produce the features I [provably] detect in plain old Redbook audio, I'm all for it. In the meantime, I enjoy the real benefits of my nearly-costless upsampling. I certainly don't anticipate putting *any* money into dedicated devices for this purpose--it would be far better spent on some Eve SC307's, among many other possible alternatives.
post #24 of 77
Quote:
Originally Posted by UltMusicSnob View Post

Holy cow, 384 kHz...I remember that number from somewhere, but not from a PCM rate, I don't think.
The science on Nyquist relationship of sample rate to freq is solid, but it's a mistake, and an oversimplification by far, to think that A/D and D/A engineering, the science of acoustics, and the science of human hearing and mental processing is all complete. Far from it. The unavoidable D/A process alone, at any sample rate, involves tradeoffs among multiple parameters. If someone figures out how to produce the features I [provably] detect in plain old Redbook audio, I'm all for it. In the meantime, I enjoy the real benefits of my nearly-costless upsampling. I certainly don't anticipate putting *any* money into dedicated devices for this purpose--it would be far better spent on some Eve SC307's, among many other possible alternatives.

I think you are being a little bit unfairly dismissive of the scientific understanding of digital and analog audio. In the grand scheme of things, audio signals are pretty simple (one dimensional functions of time), and our ability to represent and reproduce them in digital formats far exceeds our auditory ability to discern the finest levels of detail. When you play the same song twice with the same reproduction chain, you hear the same thing twice---the hardware out-performs your auditory system.

 

You cannot conclude that because based on the limited information you've given us over an internet forum, that because us random collection of audio enthusiasts cannot explain to your satisfaction the mechanism by which you are successfully ABXing the same source 16/44 PCM stream from it's digitally upsampled 16/192 PCM stream, that the actual explanation is beyond the capabilities of science.

 

I assure you, there a reason why you can ABX the CD quality audio from the upsampled version, and if someone was able to ask the right questions and take the appropriate measurements, it would be quite simple to determine that reason.

 

This is a simple analog signal we're talking about, this isn't quantum gravity, turbulence, or the Reimann hypothesis.

 

I, for one, think that it's incredibly neat that you can so successfully ABX 44 and 192k audio streams of the exact same source data. The implication that you are audibly picking up on cues that exists somewhere north of 20kHz is fascinating, which is why i've been following this thread. I'm dying to see what the waveforms being sent to your headphones look like for each of the sample rates.

 

Cheers!

post #25 of 77
Quote:
Originally Posted by ab initio View Post

I think you are being a little bit unfairly dismissive of the scientific understanding of digital and analog audio. In the grand scheme of things, audio signals are pretty simple (one dimensional functions of time), and our ability to represent and reproduce them in digital formats far exceeds our auditory ability to discern the finest levels of detail. When you play the same song twice with the same reproduction chain, you hear the same thing twice---the hardware out-performs your auditory system.

 

I plead limited space to be clear!redface.gifOnce again I've made too much of a single point. I'm by no means trying to dismiss science and claim "It's a Mystery!"  Instead, I'm reading the still-developing science in the area, such as Kunchur 2008, van Maanen 2010, I'm particularly interested in cutting edge empirical work on temporal resolution, such as Oppenheim and Magnasco 2013 (this one's behind a wall, I'm accessing with faculty privileges):

"We have conducted the first direct psychoacoustical test
of the Fourier uncertainty principle in human hearing, by
measuring simultaneous temporal and frequency discrimi-
nation. Our data indicate that human subjects often beat the
bound prescribed by the uncertainty theorem, by factors in
excess of 10. This is sometimes accomplished by an
increase in frequency acuity, but by and large it is temporal
acuity that is increased and largely responsible for these
gains. Our data further indicate subject acuity is just as
good for a notelike amplitude envelope as for the Gaussian,
even though theoretically the uncertainty product is
increased for such waveforms."

 

It's important that an evidence-based result here calls into question a limit that was formerly assumed on the basis of a mathematical construction.

 

You cannot conclude that because based on the limited information you've given us over an internet forum, that because us random collection of audio enthusiasts cannot explain to your satisfaction the mechanism by which you are successfully ABXing the same source 16/44 PCM stream from it's digitally upsampled 16/192 PCM stream, that the actual explanation is beyond the capabilities of science.

Wow, I would never claim that, sorry for the misrepresentation. I am *absolutely* sure that it IS well within the capabilities of science. What I'm calling very much into question is simplistic claims like "When you upsample, there is absolutely no difference whatsoever, unless the program you use is introducing artifacts. If you can hear a difference, it isn't music you are hearing. It's noise introduced by your upsample conversion." There is a LOT more to music, its capture, storage and reproduction, than is signified in reductive claims like this. The time domain of sound and psychoacoustics is *critical*, and far from settled as a large set of open scientific questions. Of course, depending on what "artifacts" is used to mean, I might be entirely in agreement with the statement above. For example, "artifacts" might mean "the time-shifting and blurring effects of D/A processing", in which case I would myself be thinking about just precisely those features.

 

I assure you, there a reason why you can ABX the CD quality audio from the upsampled version, and if someone was able to ask the right questions and take the appropriate measurements, it would be quite simple to determine that reason.

Yes, it always comes down to that, the right questions and appropriate measurements--no argument at all there.

 

This is a simple analog signal we're talking about, this isn't quantum gravity, turbulence, or the Reimann hypothesis.

Hmm. Analog, yes, simple not so much. It's a stereo signal with a complex waveform, decoded by mental processes into multiple spatial components, subject to many stages of processing between original creation and final listening. To me, it's far from simple. I *don't* mean mysterious or unknowable. I *do* mean complex and not fully understood.

 

I, for one, think that it's incredibly neat that you can so successfully ABX 44 and 192k audio streams of the exact same source data. The implication that you are audibly picking up on cues that exists somewhere north of 20kHz is fascinating, which is why i've been following this thread. I'm dying to see what the waveforms being sent to your headphones look like for each of the sample rates.

Again, apologies for the implication. I'm sure I am NOT picking up cues north of 20kHz, at least not in any way that's available to me consciously for the purposes of deciding a trial. My hearing, far as I can tell, is limited somewhere under 18 kHz. My success in detection has to lie elsewhere. Prime candidates I would say are D/A effects in the audible band due to filter interactions, and temporal resolution cues.

 

Here's something *really* telling to me, from Oppenheim: "

We further found that composers and conductors
achieved the best results in task 5, consistently beating
the uncertainty principle by factors of 2 or more, whereas
performers were more likely to beat it only by a few

percentage points."

 

Turns out, I *am* a composer, of contemporary classical music, with a Doctoral degree, and years of conducting experience. Listening into the music for the cues I'm trying to find IS very much like listening into the orchestra to hear that the second oboe was flat in the 17th bar.

 

Cheers!

post #26 of 77
Quote:
Originally Posted by UltMusicSnob View Post
I plead limited space to be clear!redface.gifOnce again I've made too much of a single point. I'm by no means trying to dismiss science and claim "It's a Mystery!"  Instead, I'm reading the still-developing science in the area, such as Kunchur 2008van Maanen 2010, I'm particularly interested in cutting edge empirical work on temporal resolution, such as Oppenheim and Magnasco 2013 (this one's behind a wall, I'm accessing with faculty privileges):
 

Yeah, that's the now infamous "Human Hearing Beats the Fourier Uncertainty Principle" paper.  The title is mostly what's wrong with it, it's quite misleading and sensational.  The paper sheds no new light, however, and contrary to what the title seems to imply, the paper does not "prove" that humans can hear things that can't be measured.  The guys a hydrogenaudio have pretty much cleared this up already.

 

Quote:
Originally Posted by UltMusicSnob View Post

Wow, I would never claim that, sorry for the misrepresentation. I am *absolutely* sure that it IS well within the capabilities of science. What I'm calling very much into question is simplistic claims like "When you upsample, there is absolutely no difference whatsoever, unless the program you use is introducing artifacts. If you can hear a difference, it isn't music you are hearing. It's noise introduced by your upsample conversion." There is a LOT more to music, its capture, storage and reproduction, than is signified in reductive claims like this. The time domain of sound and psychoacoustics is *critical*, and far from settled as a large set of open scientific questions. Of course, depending on what "artifacts" is used to mean, I might be entirely in agreement with the statement above. For example, "artifacts" might mean "the time-shifting and blurring effects of D/A processing", in which case I would myself be thinking about just precisely those features.

 

 

As I think about how to respond, I've decided to simply ask what you mean by "time domain of sound", and how you're applying it to D/A processing, resulting in "time-shifting and blurring". We can have infinite linear time delay (think long term storage), where data T is constant over the audio band, or non-linear time delay (group delay) where delta T changes with frequency, which becomes generally audible at some point if there's enough change at frequencies where harmonics are still in the audible band.  I assume you mean the latter, but I'm questioning your application to D/A processing and up-sampling.  Clarify?

 

Quote:

Originally Posted by UltMusicSnob View Post
Hmm. Analog, yes, simple not so much. It's a stereo signal with a complex waveform, decoded by mental processes into multiple spatial components, subject to many stages of processing between original creation and final listening. To me, it's far from simple. I *don't* mean mysterious or unknowable. I *do* mean complex and not fully understood.

 

"Stereo" is actually two complex waveforms with complex timing information contained as differential information, but the recording an electrical representation of a amplified microphone's signal today is done routinely with far greater precision than ever before.  The implication, still, is that there is something not fully understood that prevents ultimate fidelity.  I would agree in the absolute, but scale that against the reality that in most cases, mic through production to release, even at 16/44.1, the resulting signal quality well exceeds practical reproduction.  Once that's true, the elimination of perception bias becomes absolutely essential in our observations, we we try to evaluate exactly how much our "observations" are of something that is "not fully understood".

Quote:

Originally Posted by UltMusicSnob View Post
I'm sure I am NOT picking up cues north of 20kHz, at least not in any way that's available to me consciously for the purposes of deciding a trial. My hearing, far as I can tell, is limited somewhere under 18 kHz. My success in detection has to lie elsewhere. Prime candidates I would say are D/A effects in the audible band due to filter interactions, and temporal resolution cues.

 

If we were still talking about 7th order elliptical, Chebychev, or even Butterworth filters at .5 Nyquist, I would agree there could be in-band effects.  But we're not talking about that, haven't been for quite a few years.  Most of the AES papers about group-delay audibility were written in the mid 1980s when those filters were a problem.  You can easily plot group delay of today's filters.  

 

I'm still waiting for the definition of a "temporal resolution cue", and how it applies to D/A processing. 

Quote:

Originally Posted by UltMusicSnob View Post

Here's something *really* telling to me, from Oppenheim: "

We further found that composers and conductors
achieved the best results in task 5, consistently beating
the uncertainty principle by factors of 2 or more, whereas
performers were more likely to beat it only by a few

percentage points."

 

 

One thing the paper did show was that trained listeners achieve better scores.  Not a big surprise, and not new information.  However, their comparison to generalized "Fourier Uncertainty" was, essentially, magnifying something already know to a sensational extent. 

 

Quote:

Originally Posted by UltMusicSnob View Post

 

Turns out, I *am* a composer, of contemporary classical music, with a Doctoral degree, and years of conducting experience. Listening into the music for the cues I'm trying to find IS very much like listening into the orchestra to hear that the second oboe was flat in the 17th bar.

Interesting, if irrelevant analogy.  The ability of a trained music professional to discern pitch with precision beyond the general populace is well documented.  However, there is also a wide range of that ability within the untrained populace, which I find far more fascinating.  Pro's aren't the only ones with precision pitch.  But that's not to denigrate your abilities at all, it's an honor to have a classical composer here!

 

I agree, though. Pitchy oboes are painful, though pitchy cellos are what cross my eyes.  

 

I would advise extreme caution when drawing conclusions based on auditioning up-sampled files against their original.  True ABX testing is very difficult to do because of hardware limitations.  It's also impossible to isolate the effects of up-sampling from the effects of DACs being asked to perform an entirely different task in terms of filtering, the specific up-sampling algorithm, etc.  Up-sampling doesn't add any information, of course, so to tune in on what the "difference" is, we'd need to know a whole lot about the DAC, how it operates, what changes with a rate change, and most important, accomplish a DBT.  That would take two identical sets of hardware, playback sync (the hard part), and a true ABX comparator.  If we don't do the test that way, the observations are so polluted with bias that there can be no reliable conclusion.

post #27 of 77
I think it's safe to assume that if the upsample sounds noticeably different from the master 16/44.1, then the upsample has been distorted in some way. Then it's just a matter of determining whether the distortion was introduced in conversion, or whether it's an equipment problem during playback.
post #28 of 77
Quote:
Originally Posted by bigshot View Post

I think it's safe to assume that if the upsample sounds noticeably different from the master 16/44.1, then the upsample has been distorted in some way. Then it's just a matter of determining whether the distortion was introduced in conversion, or whether it's an equipment problem during playback.


But it's not necessarily a detrimental effect ("distortion") from the original 44.1 kHz sample file. Thing is, 44.1 digital data has to be turned back into an analog electrical Line Out, and there are multiple parameters to address in that process. I guess what I'm saying is that ALL digital files undergo changes due to processing before they get to your ears. And we already know that early CD Redbook implementations ran into mistakes on both the parameter choices and the quality of the components involved.

 

Even if the 44.1 kHz file were as near a perfect representation of the original acoustic signal as it's possible to get (I haven't detected anything in ABX that's purely electronic-sourced, btw, my test file has to have live acoustically generated elements), there are going to be effects of processing getting that all the way from bits on a hard drive out to my ears.

 

And the D/A of 44.1 and 192 is not the same. There are parameters and trade-offs **in both**. Question is, which set of processing effects--present in ALL files--provides the best end product. I have an answer: 192 at 24 bits (I seriously doubt the bits have much to do with it in most cases). Here is some highly rigorous work on the question--the main points being that all signals from 4 kHz to 384 kHz and beyond are transformed in DAC, and that it's not just about Nyquist and freq content--timing also matters. http://www.physics.sc.edu/kunchur/align.pdf


Edited by UltMusicSnob - 8/20/13 at 10:08am
post #29 of 77
Quote:
Originally Posted by UltMusicSnob View Post


But it's not necessarily a detrimental effect ("distortion") from the original 44.1 kHz sample file.

You might not find it to be detrimental, but it is definately a distortion. Any change from the original is distortion, so by definition, any detectable difference from the master file is a distortion (or perhaps an imbalance of some sort).
post #30 of 77
Quote:
Originally Posted by bigshot View Post


You might not find it to be detrimental, but it is definately a distortion. Any change from the original is distortion, so by definition, any detectable difference from the master file is a distortion (or perhaps an imbalance of some sort).


The point to bring on board is that 44.1 kHz digital data **also** undergoes changes in the path from bits to sound waves. It's not a question of with or without. 44.1 has trade-offs and multiple parameters just like any other digital file, when processed back to analog.

 

It's odd for me to use the term "distortion", when the effect in question could be "provides improved phase alignment after D/A, when compared to a lower sample rate". "Distortion" implies a departure from a preferred norm, but in this case the processing involved could be moving from "norm" to "new, more-preferred". I am suspending an assumption which seems to be made a lot, which is that 44.1 kHz already represents a quality cap ***in all respects***. The science (see the papers cited above) does not support that.

New Posts  All Forums:Forum Nav:
  Return Home
  Back to Forum: Sound Science
Head-Fi.org › Forums › Equipment Forums › Sound Science › Can someone explain audio resolution to me?