An interesting thread where Paul Frindle, an EE working in digital audio for a long time, talks about his experiences with audibility, measurements, problem investigations, and solutions. Searching for the term 'dither' on each page brings up some interesting discussion. Page 5 has quite a bit about one such problem investigation. As much of the thread as I have read so far is reasonably technical, yet described in plain language. Paul Frindle - Is This Truth Or Myth? - - Page 5 - Gearspace.com
Okay, a quote from Mr. Frindle:
"The best way of finding out if we do actually hear something (rather than just thinking we do) is the ABX test. This is a purely differential test where one has to try and tell whether X is the same as A or B - where X is randomly assigned to one or the other at the beginning of each test run.
In this way we have to put our bias and subjective impressions to one side and just listen for any differences, regardless of what they might be. If you add up a good few runs of this you can quickly find out if anything can be heard reliably. I used to do this myself as part of the design process - and it's horribly revealing, in that much of what you thought you could hear, you weren't really actually hearing - and much of what you thought you could never ever hear, turns out that you actually can against all expectation!!
It was results based on this test over a decade or more than have formed my opinions about how sensitive we really are in ways that are not well documented - and seem absurd in a practical and logic sense..
At the point where you can reliably hear distortion artefacts that cannot even be measured on the world's most sensitive test kit - it seems you are launched into a pretty undefined world, where the listening test result outweighs the measured performance results. The dither corellation issue fell into this area fair and square.
And of course these results can seem absurd - but they could be reproduced any day using the ABX test - even the most critical had to admit they heard it.
Of course this has nothing to do with what you might prefer to hear - it only tells you what you can hear.
ON the subject of day to day location; this is very very important. Many of the artefacts could only be heard in a fairly normally reflective environment, because they rely on our (apparently amazing) ability to compare things with the room responses around us (we are after all normally immersed in such environments and it's not surprising we are most sensitive within these).
Also it has to be said that although these differences could be heard with almost any monitors, some were much more effective where trying to guess what the causes were. For instance the dither thing was readily audible in the guys mastering suite - but there was no way one could guess the cause there. That only occurred when I was back at the lab listening with electrostatic speakers for hours on end. The thing is that trying to detect the cause by totally honest and open minded listening is a deep discipline you have to acquire - simply because the actual causes can be so completely unbelievable before they are found you would never guess them from logic and reason."
"The best way of finding out if we do actually hear something (rather than just thinking we do) is the ABX test. This is a purely differential test where one has to try and tell whether X is the same as A or B - where X is randomly assigned to one or the other at the beginning of each test run.
In this way we have to put our bias and subjective impressions to one side and just listen for any differences, regardless of what they might be. If you add up a good few runs of this you can quickly find out if anything can be heard reliably. I used to do this myself as part of the design process - and it's horribly revealing, in that much of what you thought you could hear, you weren't really actually hearing - and much of what you thought you could never ever hear, turns out that you actually can against all expectation!!
It was results based on this test over a decade or more than have formed my opinions about how sensitive we really are in ways that are not well documented - and seem absurd in a practical and logic sense..
At the point where you can reliably hear distortion artefacts that cannot even be measured on the world's most sensitive test kit - it seems you are launched into a pretty undefined world, where the listening test result outweighs the measured performance results. The dither corellation issue fell into this area fair and square.
And of course these results can seem absurd - but they could be reproduced any day using the ABX test - even the most critical had to admit they heard it.
Of course this has nothing to do with what you might prefer to hear - it only tells you what you can hear.
ON the subject of day to day location; this is very very important. Many of the artefacts could only be heard in a fairly normally reflective environment, because they rely on our (apparently amazing) ability to compare things with the room responses around us (we are after all normally immersed in such environments and it's not surprising we are most sensitive within these).
Also it has to be said that although these differences could be heard with almost any monitors, some were much more effective where trying to guess what the causes were. For instance the dither thing was readily audible in the guys mastering suite - but there was no way one could guess the cause there. That only occurred when I was back at the lab listening with electrostatic speakers for hours on end. The thing is that trying to detect the cause by totally honest and open minded listening is a deep discipline you have to acquire - simply because the actual causes can be so completely unbelievable before they are found you would never guess them from logic and reason."
From my perspective, some types of sort-of gritty sounding noise sounds a lot like distortion.
Obviously if something doesn't show up on a FFT, then it is not time-invariant HD.
Perceptually speaking, people tend to describe new things using words with similar preexisting associations. If the brain associates a certain perceptual 'grit' sound with distortion (maybe high order HD in this case), then shouldn't be surprising someone describes something they hear as a 'distortion.' I don't think they necessarily mean a mathematically defined type of 'distortion.'
Obviously if something doesn't show up on a FFT, then it is not time-invariant HD.
Perceptually speaking, people tend to describe new things using words with similar preexisting associations. If the brain associates a certain perceptual 'grit' sound with distortion (maybe high order HD in this case), then shouldn't be surprising someone describes something they hear as a 'distortion.' I don't think they necessarily mean a mathematically defined type of 'distortion.'
Last edited:
To further my comments a couple posts above, Bruno Putzeys wrote about a claimed audible phenomenon which he dubbed hysteresis distortion, not hysteresis noise. That being so despite it not showing up on as a spur on an FFT. Why? Seems likely to me it was because since the noise-like effect is correlated with an audio signal, it sounds more like a distortion rather than a noise.
OTOH, the sigma-delta modulator guys have chosen to characterize a different type of signal correlated noise as 'noise floor modulation.'
Call it what you want. If the term 'distortion' is going to be appropriate in some cases then maybe we should consider rethinking the conventional dogma that there is only FFT-based linear distortion verses FFT-based non-linear distortion.
As a further consideration, not all FFT distortion measurements taken in the area of audio are of completely non-time invariant phenomena. They may be time-invariant only in approximation.
Why does it matter? Why should we consider using more complex models of audio device performance? I would say its because of the decades old unsolved dispute between the so-called subjectivists and objectivists. I believe there is more to story than old, ineptly conducted ABX tests suggest. To be fair to the old researchers, nobody knew then as much as is now known about the difficulties of reliable human perceptual testing.
OTOH, the sigma-delta modulator guys have chosen to characterize a different type of signal correlated noise as 'noise floor modulation.'
Call it what you want. If the term 'distortion' is going to be appropriate in some cases then maybe we should consider rethinking the conventional dogma that there is only FFT-based linear distortion verses FFT-based non-linear distortion.
As a further consideration, not all FFT distortion measurements taken in the area of audio are of completely non-time invariant phenomena. They may be time-invariant only in approximation.
Why does it matter? Why should we consider using more complex models of audio device performance? I would say its because of the decades old unsolved dispute between the so-called subjectivists and objectivists. I believe there is more to story than old, ineptly conducted ABX tests suggest. To be fair to the old researchers, nobody knew then as much as is now known about the difficulties of reliable human perceptual testing.
Last edited:
He's a big believer in blind testing. That's a good thing. I have written about doing it myself before, and recommended that everyone should do it to help stay 'calibrated' as to what is real and what isn't. I still do some of that with my listening panel. However, I have also written about what should be fixed in foobar ABX to make it more useable.
Last edited:
We should keep in mind, that these were posted in 2010 and were reflecting a lot of the difficulties Frindle and his team were facing while developing the Sony Oxford console .
But he is a good example for keeping an open mind while otoh trying to find corrobation by using the scientific toolbox. That was the reason why I quite often cited his AES-presentation from 1997 and some of his posts about various dither phenomenons as well.
And I'd assume that he did not mean it as "could never be measured" just as "measurement gear could not measure it at the time".
In the first post on page 4 of the thread he wrote about the "dither-thing":
Paul Frindle - Is This Truth Or Myth? - - Page 4 - Gearspace.com
Last edited:
20 years ago I built a system to facilitate a quick self administered blind A/B test, that works quite well for many circumstances. Basically it is an A/B selector that starts from a random choice of A or B. The two devices under test are now unknown and hence X & Y. The listener can select X or Y to their heart's content without knowing whether X is A or B, and whether Y is B or A. Once an artefact or character difference is heard, the listener needs to un-blind X & Y to find out what process to attribute the observation too. I have in the past conducted double blind tests to AES recommended practice and also A/B/X tests, and I find this an equally valid but much more rapid way to determine by oneself the absence or presence of audible differences than A/B/X or DB tests. It is also a very good training system to overcome expectation bias and build confidence in personal hearing acuity.
My personal experience is consistent with Markw4's posts. I found some artefacts are most obvious with fast X/Y switching, whilst other artefacts make take hours to hear, but once learned can be identified audibly fairly quickly. I think this experience explains to some extent the divide between objectivists and subjectivists - not everything is exposed by quick A/B comparisons.
My personal experience is consistent with Markw4's posts. I found some artefacts are most obvious with fast X/Y switching, whilst other artefacts make take hours to hear, but once learned can be identified audibly fairly quickly. I think this experience explains to some extent the divide between objectivists and subjectivists - not everything is exposed by quick A/B comparisons.
Last edited:
Maybe, but before engaging, would you please explain your understanding of "FFT-based linear distortion" and "FFT-based non-linear distortion" concepts? I must have missed this class in Signal Processing.
Unless it's yet another exercise of pulling buzzwords from a hat and putting them together, hoping somebody will be in ave after reading
In 2010 the signal processing knowledge certainly existed to measure most if not all that has been shown. Nice to see Mr. Frindle's honest comment about lack of controls in his experiments.
We are stuck it seems at times in, "If I can't plug my box into an AP and pick a test off of a menu it can't be measured".
What Frindle described happened most likely around 1993 - 1996.
Regarding the usage of controls - do you really think, Frindles post was intended to allow conclusion about controls in his experiments?
Indeed. Reads to me like the type of disclaimer the legal department wants to see, not necessarily what Frindle personally believes in relation to the cases he described.
The fairly direct declarative statement (mentioned by Scott Wurcer) wasn't posted by Paul Frindle.
Frindle himself (according to his posts and the AES-paper I've quoted quite frequently in the past) did a lot of blind tests even using the ABX protocol, without metioning the detailed procedure, though.
I wonder why you felt the need to reply to scott wurcer now. His account has been disabled since September of last year.