You need to think about what diffraction is. It is a sound wave that is reflected from a discontinuity as it travels. At LF the waves just pass by with no effect, but as the frequency goes up they begin to diffract depending on the physical sizes of the objects relative to the wavelengths involved. Once they begin to diffract, the amount stays pretty much constant, so it's a kind of high pass function.
We did not study the location in frequency of the diffraction (the HP point), but the amplitude and delay. The delay depends on dimensions involved, as do ALL other details. We varied the delay of a "diffraction like" signal and noted the effects of SPL and delay on audibility. Both were significant.
It would be hard to find the exact physical object that would yield the functions we used, but suffice it to say that some object would. To understand how one would perceive, in detail, any object would require an infinite amount of experiments. That's why I say this is a data point. How well it covers the surface of all data points is unknown. But we can say that any hypothesis that does not agree with this data point is likely wrong.
We did not study the location in frequency of the diffraction (the HP point), but the amplitude and delay. The delay depends on dimensions involved, as do ALL other details. We varied the delay of a "diffraction like" signal and noted the effects of SPL and delay on audibility. Both were significant.
It would be hard to find the exact physical object that would yield the functions we used, but suffice it to say that some object would. To understand how one would perceive, in detail, any object would require an infinite amount of experiments. That's why I say this is a data point. How well it covers the surface of all data points is unknown. But we can say that any hypothesis that does not agree with this data point is likely wrong.
Thanks for the explanation and the careful boundaries you put on generalizing your findings.
Granted, you demonstrated that careful listeners using fancy inserted headphones can detect that your tinkered stimuli compared A-B to non-modified sounds differ to a statistically significant degree. To your credit, you note "significant" but do not claim the factors you varied are materially meaningful since the resulting "effects" (as we say in the trade) are small.
Which aligns with other respected researchers such as Toole and Lipshitz who agree that special sounds with phase aberrations can be sensed in headphones and possibly in anechoic chambers. But they say there isn't detection with speakers in rooms, special test sounds possibly excepted (like playing square waves).
Your research did nail down why ESL users (who rarely can play awfully loud) find other speakers when demo'ed playing loud sound distorted: you showed our hearing is distorted as loudness rises.
B.
Granted, you demonstrated that careful listeners using fancy inserted headphones can detect that your tinkered stimuli compared A-B to non-modified sounds differ to a statistically significant degree. To your credit, you note "significant" but do not claim the factors you varied are materially meaningful since the resulting "effects" (as we say in the trade) are small.
Which aligns with other respected researchers such as Toole and Lipshitz who agree that special sounds with phase aberrations can be sensed in headphones and possibly in anechoic chambers. But they say there isn't detection with speakers in rooms, special test sounds possibly excepted (like playing square waves).
Your research did nail down why ESL users (who rarely can play awfully loud) find other speakers when demo'ed playing loud sound distorted: you showed our hearing is distorted as loudness rises.
B.
Last edited:
I'm not "on my own". I'm respectfully standing behind Toole and Lipshitz. I found this quote attributed to Toole:
" It turns out that, within very generous tolerances, humans are insensitive to phase shifts. Under carefully contrived circumstances, special signals auditioned in anechoic conditions, or through headphones, people have heard slight differences. However, even these limited results have failed to provide clear evidence of a 'preference' for a lack of phase shift. When auditioned in real rooms, these differences disappear.. ."
I stand behind Earl's methods 100% (and very careful work it is), but doubt the relevance to this thread. I don't know the latest research. But I believe it's possible to find a distorted test sound (or commercially-released singer, in Earl's test) that people can A-B with headphones but couldn't reliably detect with speakers and by memory, as in the real-world of DIYaudio.
B.
Last edited:
Indeed. Having faith in this data point is one of the best moves I've made in this field. It has changed my view significantly. I can hear diffractions (some at least, maybe not all), but I can.
Maybe you should read Earl's article again before standing so solidly behind his "data point". There is no "point" but an elaborate set of plots showing slight effects even if A-B'ing a special sound with fancy earphones and only when the distortions are at the higher end. His conclusions as to statistical significance arise from a complex and non-intuitive statistical processing. So it isn't clear how they map on to real-world listening.
Posts like PLB's earlier are "tribal" and that sort of invective (posting nothing but, "you are not one of us, Ben") is shameful.
After reviewing the subject in some depth, as of 2013, Rod Elliott (who is one of the gurus I find most trustworthy to read) had this to say:
"From the number of websites and articles about phase audibility, it really looks like there are people desperately trying to prove that phase and/or phase shift is audible. So far, none has succeeded - the basic assumption that we are not sensitive to phase (with real-life signals at least) holds true."
B.
Last edited:
I think that Toole, Ellito etc. are talking about large scale phase shift (in crossovers). I would be very careful to use that as an argument to baffle diffranction-induced phase shifts' audibility - which happen in very limited frequency range, as peak/dip in response.
Anyway, diffractions from driver geometry, mounting and baffle have very short delay. They are easily masked by much more audible reflections from the room. Still, these diffractions are worth of minimizing - at least the measurements look and sell better then!
Anyway, diffractions from driver geometry, mounting and baffle have very short delay. They are easily masked by much more audible reflections from the room. Still, these diffractions are worth of minimizing - at least the measurements look and sell better then!
There are more than just "phase aberrations" in Earl's test signals - look again at the amplitude response caused by the comb filtering when the delayed signal is mixed in - it's pretty obvious looking and I fail to see how it would not be fairly audible when the amplitude deviation is as large as shown in the example graph as we're talking about peaks and dips of several dB. (Only one time delay and amplitude was shown in the published graph, some of the samples would have been a lot more subtle)
The real question of the research is does this effect become more obvious and/or obnoxious as SPL increases, which the research seems to suggest is yes.
Further, regarding the audibility of such effects in a real room, keep in mind that the frequency response aberrations caused by diffraction are present in the "direct" signal, and at frequencies above about 2Khz the direct signal dominates our perception of the quality of the sound.
It is not "swamped out" by room effects. If there are sharp peaks and dips in response >2Khz caused by diffraction, we will hear them regardless of the fact that they won't be present in the overall room power response. (As the power response in the room won't tend to show diffraction effects as it's a spatial average)
I've re-read your paper and it is indeed one that I've read before - albeit over 5 years ago.
I think the delay is a factor in audibility in so much as it affects the density of the peaks and dips and the "slope" of the amplitude response variations in the summed response, with steeper slopes (in my opinion) sounding "nastier" at higher SPL than flatter, smoother responses with more gentle slopes. In other words steep, rapid changes in amplitude are to be avoided for good sound quality.
Also the specific delay will affect specific peak and dip frequencies for the first few discrete peaks and dips before they become very dense, and as we know not all frequencies are created equal when it comes to how sensitive to them we are.
In your example in Figure 2 for example, the first peak happens to fall on 3Khz, which as is well known, is the most sensitive frequency in our hearing system, and probably the frequency where our hearing is most easily "irritated". Any peaks in this frequency range may make imaging seem hyper-real and vivid but quickly become fatiguing and irritating if you turn up the SPL.
So sliding the delay time may have a secondary effect of hitting more and less "sensitive" or irritable portions of our hearing range, so this possibility needs to be considered in evaluating the results.
On the whole response slope issue, I've never bought into the "High Q resonances are less audible and thus more desirable than Low Q resonances" that Toole among others subscribe to.
Yes it's true that from a tonal balance perspective a Low Q resonance is more audible than a High Q resonance of the same amplitude, because the error in frequency response is over a wider range and has more "area under the graph", and this is probably what they are thinking of.
But tonal balance and sound quality are not the same thing even though frequency response plays a role in both.
You can have a speaker with good tonal balance by way of a balanced 1/3rd octave averaged response that still sounds nasty and harsh due to narrow band high Q resonances or diffraction that cause steep, spikey frequency response variations, and conversely you can have a speaker that is a bit tonally imbalanced that sounds much cleaner to listen to because even though it has errors in the frequency response they are very gradual and smooth errors that only result in a tonal imbalance.
For a long time I've felt that a large part of sound quality comes from the smoothness of the response, not the overall tonal balance, and this is especially so at frequencies above about 2Khz where our ears start to become more easily irritable.
This is why I always look at the un-smoothed narrowband response of a speaker and specifically look for any sharp peaks/dips or rapid changes in response. Apart from gauging overall tonal balance I find 1/3rd Octave smoothed measurements almost worthless at judging the quality of a speaker, which is why I don't pay any attention to them.
Because high Q resonances are narrow, it's possible for them to "slip between the cracks" in certain songs or styles of music that tend to have fairly discrete spectra as they may not always get excited, whereas low Q resonances will always tend to get excited, and this may also be a reason why Toole etc say that Low Q resonances are more audible.
However when music does excite a high Q resonance the result can be very nasty, and this often gets blamed on the recording because "most other recordings sound fine".
I've had very specific pieces of music that for years I've thought were badly recorded and a bit "edgy", and sounded that way on multiple different speakers of different design, however when I've really put in the work to try to eliminate all the resonances >2Khz with a fairly elaborate crossover/EQ setup and driver damping mods, much to my surprise nearly all of those recordings which I was blaming for sounding edgey for years now actually sound fine. In one case there was a 1dB peak at 7.5Khz, approximately 1/4th octave wide on an otherwise pretty flat treble response - correcting that small peak was enough to remove the edginess on many recordings. A small peak that would probably be considered acceptable in most designs and ignored.
The specific recordings are still "pushy" in the relevant frequency range but no longer edgy. The recording just happened to have strong spectral content in a frequency range where a lot of speakers have problems with resonances and due to that the recording gets the blame.
Certain kinds of music however do seem to excite all high frequency resonances quite readily - in particular massed orchestral strings and heavily distorted electric guitar, both of which will immediately show up any resonances in the upper midrange through to the treble, presumably because the spectral distribution of these kinds of instruments is so rich and dense.
Even though I'm not a fan of classical music by any stretch, I do have a few pieces of good quality orchestral music that are rich in stringed instruments, and they are an excellent test of high frequency resonances - if you can turn up the volume on these songs without any edginess or desire to turn the volume back down again your speakers are doing well!
I would suggest that a piece of classical orchestral music rich in strings would have been a better choice for your test than the Tracey Chapman used.
Last edited:
"Phase shifts" and diffraction delay (like we studied) are completely different things.
Of course you can assume that our results don't apply, but it is just an assumption on your part since you have not shown that they don't either.
You see Ben, you simply do not understand that phase shifts are not the issue. I agree with the above assertion, but it has nothing to do with what we studied.
I would have agreed with you on this except when I went back and reread Floyd's claim. In it he compares resonances of equal height, NOT equal energy, The lower Q curve of equal height has much more energy than the high Q.
Sean Olive once commented that Tracy Chapman recordings showed the best resolution for listener discrimination of the recordings that he tested. That's why we used it.
Funny that you misunderstand the logic.
As Juhazi (once again very helpfully) points out, you'd expect big, easy to spot problems. But Geddes and Lee showed the effect is quite small unless you're boosting the distortion and even when tested under highly revealing methods (earphones and A-B comparison).
(Finding "statistical significance" means journal editors will accept for publication. It does not mean the phenomenon amounts to much. And recent evidence and statistics re-thinking suggest some "significant" research may not be replicable either.)
Odd that this thread seems hell-bent on fixing a problem that authorities, Earl except, doesn't seem to think happens or maybe happens to an undetectable degree in the diffraction scale we are talking about, DBMandrake excepted. Like with comb filtering and other phenomena that people "hear" only after seeing a physics diagram, there's a big difference between having unmitigated faith in physics versus having some grasp of how the "meat computer" works.
I do not mindlessly apply that criticism to all acoustic anomalies. The acoustical reality of speaker Doppler distortion may be present as a human perception. Likewise, there are time and group delay issues that you can hear when a threshold is exceeded.
Earl's research is a good example of the kind of human research that is needed. But you have understand what he found and where that finding applies, as he pointed out earlier.
B.
Last edited:
I’d have to disagree. Anecdotal, but comparison between our CGR rectangular monkey coffin box, and the same alignment in a box with a crude teardrop plan section with big chamfers on the sides, shows that the 2nd box has the capability to disappear to a greater extent, ie it has a lower diffraction signature. Something of significant importance if you are looking for the best image/soundstage capability.
dave
I don't think we're actually disagreeing here.
I did specifically mention that Toole's comment about lower Q resonances being more audible than higher Q ones is based on an equal amplitude, which is an unnatural comparison for mechanical resonances of different Q.
In this equal amplitude scenario the lower Q resonance is indeed more obvious as a tonal imbalance (as you say there is more energy in it) however I would argue the high Q resonance when excited with the right signal is more objectionable sounding, typically sounding edgy or harsh if it is a high frequency.
Toole's comments about low Q vs high Q resonances, because they are based on equal amplitude seem to be frequently taken out of context or misrepresented with people drawing the conclusion that you shouldn't bother trying to damp a high Q resonance in a driver or cabinet as the lower Q resonance that results will be more audible. Not true!
I remember speaker dave pointing out in a thread a few years ago that any high Q mechanical resonance which has damping applied will drop in amplitude at a rate in proportion to its increasing width that makes it less audible overall.
Last edited:
What "authorities" do not believe that diffraction is an issue. Toole hardly ever mentions it, but that's not like saying that he does not believe that it is important. And all the quotes that you showed earlier don't apply to this discussion at all. In the "big picture" diffraction is down the list (when compared to frequency response and polar response,) but certainly higher IMO than any nonlinear distortions.
Simon - I agree that the level of the aberrations was high, but there is a very good reason for that. We were forced to test at what has to be considered as very low levels - no greater than 80 dB in the ear. That's because we were using public university facilities and the restrictions placed on human testing are extremely severe without very complicated special consideration. Since the audiblity increases with level, normal signal levels of 90+ dB (these are all "C" weighted) would have much lower levels of the aberrations audible. In fact, based on our results, I would estimate that at signals nearing on 100 dB (C) these aberrations would dominate the perception, even at low aberation levels.
We're absolutely NOT disagreeing! Do you think that every comment that I make is a disagreement? I agree with you about the audibility of high Q resonances, I was just trying to explain why Toole's comment has to be considered in light of how they did the tests.
I've always meant to work this out mathematically because I believe that it is true on a log-log curve, but probably not true on a linear-linear one. Since the energy requires a linear-linear calculation, the energy (area under the curve) would drop substantially as the damping is increased. Let's face it, it has to.
What evidence do you have that baffle diffraction is easily masked by room reflections ?
You're comparing apples and oranges. A room reflection as you point out has a much, much longer delay time than the delay of diffracted signal from a baffle edge.
So much longer in fact that the room reflection is outside of the fusing window of our hearing at high frequencies, and thus gets processed by the brain as an echo to be filtered out rather than being part of the original signal, whereas the diffracted signal is within the fusing time window, so gets summed and treated as part of the direct signal by the processing in the brain.
This is an absolutely fundamental difference. Also while a reflection of a wall like a sidewall gives a time delayed impulse that looks very much like the original but with minor modifications to frequency response, the diffraction from a baffle edge is extremely smeared in time, and does not look like a duplicate but delayed impulse with a typical rectangular cabinet.
From the point of view of our hearing, diffraction alters the direct path frequency response that we hear - especially on axis to the baffle, but has much less effect at wide off axis angles that typically result in sidewall reflections and theoretically has no effect on the power response once past the baffle step transition frequency, as for any given spot frequency diffraction has the effect of distorting the polar response of the speaker, with different frequencies being distorted in different ways.
Minimising diffraction is not just about making a graph look good, it genuinely makes a speaker sound better and also typically makes it much easier to achieve a smooth flat on axis response.
- Status
- This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.