 
|
Audio Asylum Thread Printer Get a view of an entire thread on one page |
For Sale Ads |
|
|
Audio Asylum Thread Printer Get a view of an entire thread on one page |
For Sale Ads |
70.181.185.2
In Reply to: RE: Better duck under that desk; posted by Phelonious Ponk on August 03, 2009 at 18:27:33
I'll believe you need no more humbling when it's clear you don't respond to reports of hearing differences with a statement implying it just has to be some psychological distortion, placebo, or whatever. If you respond with 'maybe so, but we can't be sure until we test..." then you have the requisite humility. I'm not holding my breath.
Follow Ups:
Well, I'm trying, riboge, I'm trying. Sometimes it's difficult not to respond that way. I've been involved in audio - pro and domestic, personal and professional, for a long, long time and I am absolutely convinced that psychological bias is one of the strongest influences on perception of quality. I have seen much of the "high-end" abandon the pursuit of high fidelity in favor of jewel boxes filled with attractive colorations (or simply nothing special). And I have seen so many audiophiles follow them down that path throwing money after "upgrades" and "hearing" ever higher and higher levels of warmth, musicality, euphonics, PRaT and other terms they and their vendors have invented to substitute for real specifications and meaningful language. When a roomful of audiophiles hear huge, paradigm-shifting improvements from something that shouldn't make any difference and can't make much difference, I'm reminded of rooms full of engineers responding to things like the Aphex Aural Exciter. And they actually DID hear a big difference. They just needed more time to understand that it was not good.
But I'm trying, riboge, I'm trying.
P
Riboge wrote:
"Maybe so, but we can't be sure until we test ..."
A fair point but, though my mathematics would struggle, surely it's possible for those more numerate than me to say unequivocally whether the changes Kuncher has detected can or can't be resolved by 44.1 KHz sampling.
If they can, then it is a question of testing whether, in practice, they are. If they can't, there seems little purpose in testing anything though no doubt someone will demand it.
On other forums where the issue has been discussed (such as HydrogenAudio and Stereophile, with the latter as unpleasant as Werner reports) there are several posts saying "This man is wrong", "Call the firing squad", "Who does he think he is?" and all the usual tosh but I can't find any explaining why he is wrong. It seems to be taken as self-evident.
Any takers here? Did I miss something?
If the sums do suggest he is wrong, it should be possible, using his apparatus, to record test tones at 44.1 and, say, 192 KHz and see what listeners can or cannot detect at either sampling rate. Good Grief! - you could even use a double-blind test (assuming you knew how to design it).
BTW, I don't get Tony's argument that the tests should have used a frequency more in line with RBCD parameters. As I see it, Kunchur was researching human hearing, not RBCD technology, and presumably (and certainly should have) chose his tones accordingly.
That said, I agree that using more than one frequency could have provided useful data with the obvious hypothesis being that discrimination would tend to be more acute in the mid-band and taper off at the extremes. How feasible it would have been in practice I can't say.
Dave
" BTW, I don't get Tony's argument that the tests should have used a frequency more in line with RBCD parameters. As I see it, Kunchur was researching human hearing, not RBCD technology, and presumably (and certainly should have) chose his tones accordingly."
I made my comment in response to an AA poster, not as a criticism of Kunchur's work. As you correctly stated, Kunchur designed his experiments and reported his results to a scientific community interested in human hearing. Had he been an engineer rather than a scientist or had he been publishing in JAES then perhaps he would have designed the tests differently.
Kunchur's choice of a 7000 kHz square wave was no error. But he did make an error by jumping into a snake pit on the two forums. I doubt that he will repeat that mistake! There is little point in debating with a bunch of characters who criticize statements taken out of context and then go on to admit they haven't even read the scientific papers that were the substance of the work. Some of the people on those forums have the smarts and knowledge to appreciate Kunchur's work, but unfortunately they appear to be suffering from overpowering egos that prevent proper use of their mental faculties.
Tony Lauck
"Diversity is the law of nature; no two entities in this universe are uniform." - P.R. Sarkar
"BTW, I don't get Tony's argument that the tests should have used a frequency more in line with RBCD parameters. As I see it, Kunchur was researching human hearing, not RBCD technology, and presumably (and certainly should have) chose his tones accordingly."I agree and would add that the only real problem here is concluding what humans can hear when listening to music, based on a study of the theoretical limits of human hearing when listening to test tones.
P
Edits: 08/04/09 08/04/09
To further Tony's point, let me add that besides the breadth of what can be considered music in the continuum from a single tone to highly complex combinations of tones and rhythms, is the issue of training. While the complexity and possible aesthetic 'divertingness' of music might make it impossible for the untrained to do with music what they do with test tones, training might very well enable listeners to overcome this and eventually make the discriminations that tests with simple tones more readily demonstrate. That is one of the great shortcomings of most feasible double blind testing in this area, it is very difficult to control for the level of training and sophistication of the subjects. This is because we are not seeking to determine just what people DO distinguish but more importantly what they CAN distinguish. The maximum of what people can distinguish in music is only manifest with the most expensive training/experience along with the most innate talent for it. Each level of experience and of talent should be expected to have different results, no?
Sorry, I don't get it. I can see the point in training myself to recognize middle C, to hear and identify complex harmonies, to understand the pattern of a composition and be able to learn it by ear. I really don't see the point in training myself to hear the the otherwise inaudible limitations of recordings when I could, instead, be listening to the music. And if I did see the point and did find the time to train myself to hear such things, I don't know how that could possibly do anything to enhance my enjoyment of music. In fact, I think it would have the opposite effect.
And you're wrong about AB/X. Of course you can test the trained; it is no less feasible than testing the untrained and has been done often. The study I referred to in this thread used audiophiles, professional recording engineers, recording engineering students, and civilians. Sean Olive's studies at HK often use a similar cross section of listeners.
P
It's not that you SHOULD train yourself thus, it's that many audiophiles DO to one degree or another and as a result can distinguish what others don't. Some studies may have used more experienced people but that is a far cry from determining how trained each is and matching result to level of training. The point is that you doubt people distinguish with what you consider music what is distinguished with test tones. I say it depends on who, with what training, experience of that kind of music, etc, in addition to whether the test psychological set and conditions are conducive to it.
On a different level, you seem to foolishly claim you are the arbiter of what in the sound presented is the music and what is something else you find worthless or even negative. It's hard to see how hearing more can ever be worse than hearing less of what is presented. What you enjoy is an even more idiosyncratic matter. Some people enjoy the deafening most of all and dislike lower volumes that actually allow one to hear what's playing.
I think we're going to just have to agree to disagree. What you're talking about is, in my view, hand-picking your test subjects to pre-determine the result. And even though I believe you would fail to accomplish that goal, the goal is the antithesis of good research. And while I know you can train yourself to hear artifacts and distortions that usually go unnoticed, even by the most discriminating listeners, it doesn't just happen, it is an active process, and it is, in my opinion, the opposite of "audiophile" in the true sense of the word -- a lover of sound. It is crossing over to be a lover of gear, a lover of the technology that produces sound.
In my opinion.
And no, I don't consider myself an arbiter of what constitutes music and never implied that. We haven't discussed my opinions of music and what I think deserves that name in this thread or any other. We have only discussed what listening to physically misaligned tweeters playing square waves says about the relative quality of digital sampling rates when listening to music - type undefined - on an actual sound reproduction system.
I believe they are two very different things, that the former says little about the latter and that the introduction of actual systems, digital files and music into the testing would have completely changed the outcome, but we can agree to disagree on that as well.
P
I didn't speak of hand-picking subjects so much as noting differences among subjects or of matching the type of subject to what is being tested. If you are studying what people in general and on average hear, that's one thing. If you are testing about what people at the the upper limit can hear, then you should use subjects who are among the best at hearing the type of material you are presenting.Your distinction between square waves thru misaligned speakers and music on a 'real' system producing completely changed outcome is an hypothesis itself which hasn't been tested, and I don't see how it could be generally. First there is a lot of synthesized music that might have elements much like the square waves. That is, the distinction may be quantitative but surely undefinable qualitatively. And if you tested the difference with a square wave, you would not know how that pertained to one with other waveforms, frequencies or whatever versus music, to say nothing of how heterogeneous that latter category is. I can think of no basis for your belief in this 'distinction' other than desire to discount what is discordant to you.
But the most absurd assertion is that audiophile listening isn't or should not be an active process. Coupled with that is the non-credible claim that hearing artifacts and distortions that usually go unnoticed somehow is a different skill from hearing finer detail, subtle differences in timbre, etc, that also usually go unnoticed--especially if you try not to try to hear them which is what you seem to be advocating for audiophiles. I recall as a young music lover taking a similar position about listening to classical music. I believed bringing knowledge of music theory and music history to bear when listening interfered with a preferable direct, less active taking the music in holistically. I am taking it that this is what you are advocating in relation to trained listening with heightened perception of the finer aspects of the musical sound. The fallacy with this position of mine earlier or yours is that listening is ALWAYS AND IRREDUCIBLY AN ACTIVE INTERPRETIVE PROCESS (read Music and Memory by Snyder to find out about this.). The more you are trained and the more you know the better you do at this interpreting.
Edits: 08/06/09 08/06/09
I don't accept the distinction between music and test tones. If one can hear differences with test tones, then someone is going hear differences with music, even if it in a "Concerto for square waves and orchestra". We all have our tastes in music. :-)
Tony Lauck
"Diversity is the law of nature; no two entities in this universe are uniform." - P.R. Sarkar
I don't accept the distinction between music and test tones."
Ok then.
P
"to say unequivocally whether the changes Kuncher has detected can or can't be resolved by 44.1 KHz sampling."
Yes, they can be resolved.
The theoretical limit on the time resolution of a sampled quantised system is the sampling period divided by the number of quantisation steps (I may be off with a factor of 2).
Applications like cellphones, modems, and optical attitude sensors exploit this phenomenon.
"I don't get Tony's argument that the tests should have used a frequency more in line with RBCD parameters"
The tests were done with 7kHz square waves. Both Tony and I wonder about what would happen with 8kHz. For two reasons.
1) The third harmonic of 7k is 21k. If the auditory system is sensitive
(in some unknown way) to sound above its nominal cutoff frequency, then surely this 21k would figure heavily in the detection process. But the thing is: 21k can be passed by 44.1kHz sampled systems.
Re-doing the test with 8k would put the third at 24k, which cannot pass
a 44.1k system.
So the 7k test cannot prove (contrary to K's claim) that 44.1k is insufficient. On the other hand, if people pass an 8k test (which was not tried), and if the detection mechanism involves the third harmonic (which we don't know), then 44.1k is proven to be insufficient.
2) The first intermodulation product of 7k and 21k is 14kHz. Many people still hear 14k. The first intermod prod of 8k and 24k is 16k. Many people don't hear 16k anymore. If the auditory system uses the level and phase of this intermod product for detecting the differences experimentally tested by K, then the outcome of such test would differ drastically between 7k and 8k.
I hope this clarifies things a bit.
bring bac k dynamic range
werner wrote:I hope this clarifies things a bit.
Thanks for your note and sorry for the late reply but, no, I’m afraid it doesn’t entirely clarify things for me.
Yes, they can be resolved . . . The theoretical limit on the time resolution of a sampled quantised system is the sampling period divided by the number of quantisation steps (I may be off with a factor of 2).
Unfortunately, I cannot translate that into the “μsecs” on which Kunchur bases his argument. Can you help?
The third harmonic of 7k is 21k. If the auditory system is sensitive (in some unknown way) to sound above its nominal cutoff frequency, then surely this 21k would figure heavily in the detection process. But the thing is: 21k can be passed by 44.1kHz sampled systems.
1. Your argument seems to assume that RBCD recordings are made to (and DACs operate at) these theoretical limits. This seems optimistic: as I understand it, upper limits are, in practice, significantly lower than 22 KHz because the Nyquist filters are not, by design, “brickwall”.
In other words, are you sure that 44.1 KHz systems can, and in practice do , resolve (pass) 21 KHz? If they do, your point is fair. If they don’t, it isn’t. My guess is that they very often don’t.
2. In any case, Kunchur’s data seem to have undermined the argument already.
In his 2008 paper ( Temporal resolution . . . ), he measured the HF threshold of a group of subjects with the two best coming in at 17.7 and 17.8 KHz. The poorest, on the other hand, only got to 9.4 KHz and would thus have been struggling a bit with the 8 KHz tone, never mind overtones (see below).
In spite of that, he/she scored 100 per cent for t = 5.6 μsec, the second lowest value to provide a significant result over the group as a whole. (See Experiment 1 and Table 3.) In other words, he/she was pretty well up there with the best of them in spite of significant hearing loss. That suggests, and Kunchur’s text makes it explicit, that the results are not down simply to the audibility of harmonics.
So the 7k test cannot prove (contrary to K's claim) that 44.1k is insufficient. On the other hand, if people pass an 8k test (which was not tried), and if the detection mechanism involves the third harmonic (which we don't know), then 44.1k is proven to be insufficient.
As I read the text, Kunchur demonstrated that all the harmonics of the test tone were inaudible (Table 1 & p 597).
2) The first intermodulation product of 7k and 21k is 14kHz. Many people still hear 14k. The first intermod prod of 8k and 24k is 16k. Many people don't hear 16k anymore. If the auditory system uses the level and phase of this intermod product for detecting the differences experimentally tested by K, then the outcome of such test would differ drastically between 7k and 8k.
As I have explained, this argument does not seem to accord with the data.
If you were to say that Kunchur makes a conceptual leap when he argues that his experiment conclusively proves that 44.1 KHz sampling rates are inadequate, I’d probably agree with you. But the argument that using an 8 KHz test tone in preference to a 7 KHz one would have proved the argument unequivocally does not, it seems to me, hold much water.
And if it does, it leaves the task of explaining why so many competent observers are adamant that higher resolutions provide better reproduction. (Higher resolutions for recording is, of course, a different thing.)
Dave
Edits: 08/05/09 08/05/09
"Unfortunately, I cannot translate that into the “μsecs” on which Kunchur bases his argument"
The (theoretical) temporal resolution of a 16-bit 44.1kHz sampled system is 22us/2^16 = 346 picoseconds. In practice it is bound to be worse, but still vastly better than 5 microseconds. What this means is that if you have two waveforms that are identical except for a temporal feature of, say, 5us, that the sampled system maintains this temporal feature, even when other aspects of the waveforms, like their bandwidth, are obviously modified when passing through the sampled system. Or even simpler: take a waveform (A), then copy it with a delay of 5 us (B). Then sampled and decode them. What you end up with are two waveforms than are delayed 5us relative to each other.
Or in the specific (but absurd) case of two fast pulses separated by 5us versus one single pulse of the same total energy, the post-sampling waveforms would 1) merge the two pulses (obviously), but 2) still keep a spectral distinction between the two signals as such (so that they can be recognised).
In satellite attitude control this principle allows digital cameras (i.e. spatial sampling) to determine the position of stars with an accuracy that is much much smaller than the image sensor's pixel pitch. That's the one non-audio application I'm quite familiar with. There are others in cellphones and modems, which are after all pretty scary sampled systems, but then I'm not into RF at all...
"This seems optimistic: as I understand it, upper limits are, in practice, significantly lower than 22 KHz because the Nyquist filters are not, by design, “brickwall”."
Digital filters can be arbitrarily brickwall, especially in software sample rate convertors. But that's even not required: in ADC and DAC chips the filter type most often used (for reasons of economy) is the half-band FIR. These have a -6dB point at 22.05kHz for CD. Their response at 21kHz is between 0 and -6dB, so almost level. So yes, CD systems can and do pass 21kHz with quite some ease.
"The poorest, on the other hand, only got to 9.4 KHz and would thus have been struggling a bit with the 8 KHz tone, "
That one 9.4kHz person is indeed something of a sore thumb. I discussed that case already at Hydrogen Audio.
BTW a limit of 9.4k does not imply struggling at 8k. Hearing is pretty much on/off at these elevated frequencies: the cutoff is very abrupt.
"As I read the text, Kunchur demonstrated that all the harmonics of the test tone were inaudible (Table 1 & p 597)."
As pure tones. But fact remains that the subjects did distinguish between the two test signals, and that if the discrimination is not based on level differences of the fundamental, then it *must* be based on something else, which leads to the lowest-order non-linear distortion products of the signals involved, namely the second harmonic of the fundamental, and the first intermodulation between the fundamental and its third. That's all there could be for the ear to work with ...
"If you were to say that Kunchur makes a conceptual leap when he argues that his experiment conclusively proves that 44.1 KHz sampling rates are inadequate, I’d probably agree with you."
Well, that's what I said at the start of this thread.
bring bac k dynamic range
werner argued that:
[the changes Kunchur has detected] can be resolved [by 44.1 KHz sampling].
The theoretical limit on the time resolution of a sampled quantised system is the sampling period divided by the number of quantisation steps . . .
Again, sorry for a late reply. I’m well out of my field here and haven’t found much by way of useful background. I’m not saying you’re wrong only that what you say seems directly to contradict what Kunchur says. See e.g. his “FAQ”:
Optical example: A binary star system is imaged through a telescope with a CCD. First, there is the analog optical resolution that is available, which depends on the objective diameter, the figure (optical correctness) of the optics, and seeing (atmospheric steadiness). This optical resolution is analogous to the "analog bandwidth". Because this resolution is limited, a point source becomes spread out into a fuzzy spot with an intensity profile governed by the point spread function or (PSF).
Next we are concerned with the density of pixels in the CCD. To avoid aliasing, the pixel spacing L must be finer than the optical resolution so that the optics provides "low pass filtering". If the pixels and their separation are larger than the separation of the centers of the two star images, the two stars will not be resolved separately and will appear as a single larger merged spot. In this case the essential feature (the fact that there are two separate stars and not an oblong nebula) has been destroyed. This is usually what is meant by "resolution" or the lack of it.
The number of bits N that can differentiate shades of intensity ("vertical resolution") has little to do with this – no number of vertical bits can undo the damage. However, details of the fuzzy merger do indeed depend on N: if the star images are moved closer together, the digital data of the sampled image will be different as long as the image shift exceeds L/N. This L/N definition of resolution applies to the coding itself and not to the system's ability to resolve essential features in the signal as described above . . .
and, in the next section (Digital Audio Recording):
However this lack of temporal resolution regarding the acoustic signal transmission should not be confused with the coding resolution of the digitizer, which is given by 23 microseconds/2^16 = 346 picoseconds. This latter quantity has no direct bearing on the system's ability to separate and keep distinct two nearby peaks and hence to preserve the details of musical sounds.
[Emphasis added - DB] You, OTOH, say that The (theoretical) temporal resolution of a 16-bit 44.1kHz sampled system is 22us/2^16 = 346 picoseconds .
I accept that Kunchur seems to confuse the notion of distinct temporal events with that of temporal resolution as investigated in his experiments and that that is unhelpful - but I don’t see it as critical to the argument.
What is more clear to me is that he says that the sampling rate is the sole critical parameter whereas you say that it is the sampling rate divided by the bit depth.
The two formulae cannot both be right - and the difference (here, nearly four orders of magnitude) is the nub of the argument as to whether RBCD is "perfect sound".
Comments?
Dave
Oh, Kunchur agrees:
"details of the fuzzy merger do indeed depend on N: if the star images are moved closer together, the digital data of the sampled image will be different as long as the image shift exceeds L/N"
Which just says that depending on the ADC resolution a sampled system can encode spatio (or temporal) features smaller than the pixel pitch (or sample period).
But then come:
"the two stars will not be resolved separately and will appear as a single larger merged spot. In this case the essential feature (the fact that there are two separate stars and not an oblong nebula) has been destroyed."
"applies to the coding itself and not to the system's ability to resolve essential features in the signal as described above "
"This latter quantity has no direct bearing on the system's ability to separate and keep distinct two nearby peaks and hence to preserve the details of musical sounds."
Above three lines state as much times that according to Kunchur it is mandatory that a system preserves "essential features", and what are these features? Well: keeping two 5-us-separated impulses separate.
That's what K claims: that an audio system has to pass two impulses less than 5 us apart.
Where is the proof that such is required for audible transparancy?
There is none. All we get are the experiments followed by an "and hence ...".
By the way: the optical analogy Kunchur cites is only a valid analogy to CD audio when the pixel display is viewed from a distance so that the
angle subtended by a single pixel is slightly smaller than half of
the angular resolution of the human eye. Only this reflects fully the
case of a 22kHz-limited system with an assumed <20kHz aural bandwidth.
And you know what? Under such circumstances would the eye merge the
two stars somewhat before the camera+display would.
But I'm fed up with this. Let this thread die. If anything more comes of it then I'm sure it will be posted at Hydrogen.
bring bac k dynamic range
Post a Followup:
| FAQ |
Post a Message! |
Forgot Password? |
|
||||||||||||||
|
||||||||||||||
This post is made possible by the generous support of people like you and our sponsors:
Top of Page ]
[ Contact Us ]
[ Support/Wish List ]
[ Copyright © 2025, Audio Asylum, all rights reserved ]
