Sorry but that would not be fair to say at all. I've made adjustments far smaller than 1dB when fine tuning the voicing of a crossover, that are certainly audible, even if it takes a bit of listening time to identify them. The audibility of a 1dB change in amplitude response depends on many factors.
For a resonance a peak is more audible than a notch for the same amplitude. For both peaks and notches they become more audible the wider they are, for the same amplitude. So a 2 octave wide 1dB peak in the midrange is easily audible as colouration but a 1/3rd octave 1dB peak probably is not.
The most audible of all is a shelf - such as a sensitivity mismatch between a mid and a tweeter causing a shelf in the response at the crossover frequency. For that even half a dB can be easily recognised. This is because an amplitude error that forms a shelf affects a very wide frequency range.
Another factor is the specific frequency. We are more sensitive to frequency response errors at some frequency ranges than others, being most sensitive to response errors from approximately 500-8000Hz, and much less sensitive to errors below or above that. So a 1dB bump at 1Khz in the midrange can be easily audible while a similar amplitude and width bump at 60Hz would be inaudible.
Additionally, frequency response errors are not just about "colouration", in other words tonal balance. A lot of the research and discussion about how sensitive we are to frequency response errors focuses only on "colouration" and often uses pink noise as the test signal when testing to see what our threshold of detection is.
However this is misleading in my opinion because frequency response errors also affect imaging due to them providing false or ambiguous HRTF cues to the brain, and the frequency response errors that can spoil precise imaging are often smaller than what we can perceive as a tonal imbalance on something like pink noise. Pink noise does not reveal this imaging effect at all because pink noise doesn't have an encoded image that recorded music does.
So when I am making a subtle half dB or less adjustment in the tweeter attenuator when checking the voicing of a speaker I'm not listening for "coloration", I'm listening to the effects on the imaging, which are apparent on music but not a test signal like pink noise. When you get certain "key" frequencies in proper balance with each other the image will "lock in" and become convincing even if there are modest errors in the response elsewhere. Knowing what those various frequency pairings are and how they influence the image is part of the secret of voicing a speaker well.
Finally I do also take issue with fanatical devotion to blind A/B testing as the only method that works or matters when deciding whether something is audible or not. While it certainly has its place and I have used it to test some things, it does have a major shortcoming - some types of subtle changes in frequency response, especially those that affect imaging, are not always immediately apparent in rapid A/B switching and take some hours or even days to form an opinion on, based on a wide selection of music.
While finalising the voicing of my crossovers I've made very subtle changes, such as lifting or dropping the midrange or tweeter by 0.2dB then listening for several days before deciding that I had it right in the first place and ending up putting it back to how it originally was. If I can listen for a couple of weeks without any urge to tweak the frequency response I know I've got it right.
That could be a result of the thick, dense carpet underlay (?) shown in the picture forming a cavity resonance. A lot of the signal is just going to bounce off the edges of it rather than being absorbed. Personally I think carpet underlay is too dense for that application as it would just generate new reflection/diffraction points without much absorption. It would be as if the drivers had been mounted recessed behind the panel.
Last edited:
Shady perceptual logic there. Yes, maybe lots of people can A/B a 1 dB difference in a 1kHz tone, but only your eye looking at an FR can detect the difference in a hills-and-valleys-FR coming at you randomly from music instruments.
And just how substituting one set of hills-and-valleys-FR for another random set affects imaging, as you claim, deserves an explanation.
Yes, when you have major bumps and dips that are different than the major bumps of another system, that does colour the sound. But that's not the same as the little bit of influence you get with square corners on your box esp. in light of the Peter/Paul principle.
B.
In situations where I notice +/-1dB and am inclined at the time to move forward, where I can demonstrate that that amount of variation is due to secondary sources I am much more inclined to not accept it.
While Peter is waiting for Paul to pay him back, he is losing interest on his investments.
While Peter is waiting for Paul to pay him back, he is losing interest on his investments.
Agreed. Diffraction is a time-domain phenomenon; delayed energy, right? The temporal resolution of human hearing is around 200 microseconds, if memory serves. That is ~7 cm -- a baffle narrower than twice of that should diminish diffraction audibility?
DBMandrake said:
Ha, my room has a height of 2,66 m
I think that ribbon tweeter certainly has to do with reduced ceiling influence. But I would still recommend trying out some diffusers. I use these. They're cheap and light, you can put these up temporarily with double-sided tape.Excellent point, I've never thought about it that way.
Actually it's self-adhesive felt used in automotive sound treatment. Haven't found something better locally, everything else that's sold in craft stores seems to be synthetic fibers, not natural wool.
bentoronto said:
I'm using 2" dome mid and 3/4" tweeter, which means almost no beaming in upper mids or treble, widening the diffraction band. At 2 kHz typical 6,5" driver is already beaming, reducing diffraction, but a 2" dome does not beam up to 6 kHz.
Now you're conflating two different things. Sensitivity to absolute amplitude changes versus sensitivity to changes in frequency response. Not the same thing at all.
We are more sensitive to certain kinds of changes in frequency response of a full spectrum of sound than we are to a change in amplitude that doesn't affect frequency response.
If I did a blind test listening to a 5 second piece of music, then a few seconds gap, then another 5 seconds listening to it with a +/- 0.5dB shift in amplitude, I very much doubt that I could even perceive a difference, let alone recognise what the difference was. With a sine wave probably even less so since the amplitude of a constant 1Khz sinewave is so heavily affected by room reflections and head position.
But if I applied a 0.5dB shelf at 3Khz to raise or lower the entire treble, I could absolutely recognise this blind and identify what the change was, an increase or decrease in treble, and do so within a few seconds listening to a piece of familiar music, and even if a long gap was left between the two playback samples.
You only need to study the HRTF to understand how one set of hills and valleys can affect imaging versus another set of hills and valleys.
Frequency response errors can introduce false HRTF cues - usually they degrade the image but in very specific cases they can sometimes artificially enhance an image, but that is probably more often by luck than design in speakers where this happens. I know some of the frequency response tricks to enhance image perception but I prefer to stay as close as possible to flat rather than trying to create a "more real than real" image that is artificial.
You need to get away from thinking of only the tonal "colouration" effect of frequency response errors, and also consider how response errors can degrade an image by confusing our HRTF mechanism. It's very important to convincing sound reproduction.
Last edited:
If the diffraction events all occur within the "fusing" time period of our hearing then in theory they should only contribute to a perceived altered frequency response in the "direct" signal from the speaker. But a change in frequency response is sufficient to ruin imaging.
Furthermore the change in frequency response can vary wildly at high frequencies over small angular shifts, thus leading to instability of the image and a poor sweet spot. So minimising diffraction so that there is only a very smooth gradual shift in frequency response as you go off axis is one contributing factor in achieving stable imaging and a wide sweet spot.
Unfortunately I'm not in charge of what is and isn't allowed in the living room.Ha, my room has a height of 2,66 m
I'm lucky that my speakers are tolerated but diffusors on the ceiling or walls definitely would not be. Looks absolutely identical to the wool based carpet underlay that I use as lining in my bass reflex cabinets...
Works very well as a lining inside the cabinet but I'm still not sure about it's use on a front panel due to its high density. Ideally you would want the wave to travel a significant distance through the material on the panel and be gradually absorbed without creating a discrete reflection, but I don't see how high frequencies could travel through wool of that density at all.
Last edited:
Right. So as narrow as possible with as much of a roundover as possible. Vivid Audio Giya and Estelon XB come to mind. Quite hard for the average DIY guy to do these shapes, though. 3D printing the front baffle maybe?
Unfortunately I'm not in charge of what is and isn't allowed in the living room.
The struggle is real
Looks absolutely identical to the wool based carpet underlay that I use as lining in my bass reflex cabinets...
It's quite soft and can be easily pulled apart, but still quite dense, I agree. But other kind of felt I can buy here is really dense and solid, suited more for hat-making or putting under the feet of furniture than for any kind of acoustical purposes.
You will never understand the problem when you look at it only as an effect on the steady state frequency response. That is not the issue.
Correct, it is its effect on the direct sounds impulse response, not on the rooms steady state response as Ben believes. But there is no single number for temporal resolution because it depends on frequency. Look at the impulse responses of Gamma-tone filter sets that represent our hearing. You will see that they are very short at HFs and very long at LF. Imaging is dominantly an effect above about 1 kHz where the impulse responses are quite short. I don't recall the exact numbers, but they are in the range that you suggest.
Any argument that nobody can sense the minor re-arrangement of bumps-and-dips caused by diffraction in the as-if steady-state FR applies far more strongly to non-steady-state listening. Only your eyes looking at the plots can "hear" the difference - steady or unsteady - because your ears surely can not* unless maybe using trick test tones.
B.
*DBMandrake excepted because he says he has truly exceptional powers of perception
Last edited:
Ben, it's not solely a FR problem. It's a direct-radiation-reflection interfering with direct-radiation, that presents itself both as comb filtering of FR and smearing time delivery.
I agree the FR component may probably never be reliably discerned. But I strongly think the time component is there to hear reliably.
I agree the FR component may probably never be reliably discerned. But I strongly think the time component is there to hear reliably.
Last edited:
Here's a gedank experiment: you're sitting in your chair and a Scientist, blind to you, is adding or removing a diffraction killer as Sgt Pepper is playing.
First, I don't think anybody could reliably tell A from B. Your ears just don't work that way even if your eyeballs looking at plots do.
Second, let's say you do sense there's a colour difference, I don't think you could reliably say whether that is "adding" or "removing".
Third, you would never be able to say "adding" sounds better than "removing" except in so far as some particular systems actually might sound better with one or the other (just as some folks prefer the sound of vinyl records with their shortcomings).
Along the same lines of testing, I've spent some time exploring my sub polarity for subs not right next to the mains (which BTW, is true for lots of systems). Sure, we all know there is the proper polarity and then there's erroneous polarity. But sometimes in a given room and specific chair, the ear (and the FR) prefers the erroneous polarity.
B.
Last edited:
Ben - a lot of hypotheses but nothing offered to support them. My claims are support by the available evidence, which contradict yours. The ear most certainly does NOT hear steady state signals the same as transient ones. There is tons of data to support this, much of it my own.
"Sure, we all know there is the proper polarity and then there's erroneous polarity."
This is completely false. In my sub optimization the subs polarities are all over the place. There is no "right" and "wrong" in an absolute sense, only what is smoothest at the listening position and thereabouts.
"Sure, we all know there is the proper polarity and then there's erroneous polarity."
This is completely false. In my sub optimization the subs polarities are all over the place. There is no "right" and "wrong" in an absolute sense, only what is smoothest at the listening position and thereabouts.
This "issue" has nothing to do with the type of test ABX, etc. or the fact that the data is analyzed statistically. Both are minimal criteria for any data that I would look at. It's a problem with the experimental design. One must design the experiment so that the data variance is minimized. This may require long periods of listening, or not. It will all come out in the data. If the variance is small then the data is good, no matter how long that takes. The only real way to "known" how good your data is, is to do it scientifically. Exclude these controls and you are simply kidding yourself that you "know" what you think that you know.
I have found the same by recognising my own tolerance to pressure magnitude variations in different bands. This is possibly all that matters to me in this specific context. Has someone put this data together in this form? Are different people significantly different in this?
I think you missed my point slightly. My point was that most ABX testing promoted by proponents of blind / double blind testing is arranged to try to force the listener to draw a conclusion about the audibility of something after only a few seconds of listening to source A and B, which compared to a normal music listening session, is an unnatural situation. Repeating those short samples of A and B in random orders a few times does little to help as they're still short chopped up samples of music/test signal.
This is sufficient for many types of comparisons where differences are more overt, but I'm making the point that on some more subtle changes that are near the threshold of detection, and particularly in the case of making judgements on imaging as it relates to frequency response anomalies, it simply takes time and a variety of test signals not encountered in a typical ABX session for the difference to be unconsciously processed and recognised.
I appreciate that you're saying that an ABX test could be designed with long listening times (with long breaks) and be conducted over a period of many days, potentially with more than once piece of test music as well, this would satisfy my objections.
However this is not the norm for ABX testing which typically wants the listener to make decisions about what they are hearing quickly, with the mistaken belief that if you can't identify a difference of two 5 second samples of audio after several random selections of them that there is no audible difference.
You two may be tripping over terminology a bit here. Nobody has defined what they mean by frequency response in the discussion. Is it the axial response of the speaker towards the listener ? Steady state room response ? Some varying window combination of both that seeks to mimic our varying temporal resolution with frequency ?
I don't recall anyone saying that they were talking about lumps and bumps in the steady state room response ? I think you have jumped to that conclusion.
I don't know about Ben but when I was talking about the effects of diffraction on frequency response and their effects on imaging I was talking about the axial response from the speaker in the listeners direction - eg the direct signal. Unless the cabinet is huge diffraction is altering this direct signal within the fusing period, and in a way that differs at different angles.
gedlee said:
Thank you. So, this picture is a representative of our temporal resolution?
Here is a better picture of the differences between felt vs. non-felt on baffle. Blue line: with felt; red line: without felt, green line: difference.
Here's the difference in impulse (blue line is baffle with no felt, for reference):
2dB difference at 3Khz - that will be very audible. 3Khz is literally where our hearing is most sensitive and it is a key frequency affecting imaging.
A difference graph is good for frequency responses but a bit confusing for impulse responses. Can you post the felt and no felt impulse responses overlaid rather than a difference ?
- Status
- This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.