Hi Tony, I wasn't exactly referring to you and wouldn't consider you naive 
I was merely wondering what to "correctest" way to go about this is. To me the MMM seems like a good way to quickly EQ the listening spot towards a desired house curve for the whole setup, basically a DRC.
However, I can definitely see some problems when averaging the reflected sound, like Dave already pointed out before.
E.g. take a nasty room mode for example which might only be present at one of the positions used to avarage. Compared with all other samples it might go unnoticed and thus unequalized but will be very noticable when moving to that listening spot later.
This kind of problem could be somewhat mitigated by individual corrections or changes in speaker placement or tow-in but may never really show up on MMM measurements.
Of course one has to be careful not to overcompensate for a problem that only affects a small part of the listening area, but I would have thought one should correct problems inherent to the speaker (e.g. XO between drivers, resonances, etc.) or room (the worst modes, absorption of first reflections, etc.) individually first and apply a DRC to achieve the desired house curve later. As far as I can see, MMM only helps to reduce the time to do latter but might mask problems by not appropriately dealing with the former...
I was merely wondering what to "correctest" way to go about this is. To me the MMM seems like a good way to quickly EQ the listening spot towards a desired house curve for the whole setup, basically a DRC.
However, I can definitely see some problems when averaging the reflected sound, like Dave already pointed out before.
E.g. take a nasty room mode for example which might only be present at one of the positions used to avarage. Compared with all other samples it might go unnoticed and thus unequalized but will be very noticable when moving to that listening spot later.
This kind of problem could be somewhat mitigated by individual corrections or changes in speaker placement or tow-in but may never really show up on MMM measurements.
Of course one has to be careful not to overcompensate for a problem that only affects a small part of the listening area, but I would have thought one should correct problems inherent to the speaker (e.g. XO between drivers, resonances, etc.) or room (the worst modes, absorption of first reflections, etc.) individually first and apply a DRC to achieve the desired house curve later. As far as I can see, MMM only helps to reduce the time to do latter but might mask problems by not appropriately dealing with the former...
Ah yes. That should work. I sometimes use that method when locked timing measurements are not possible (as when using a USB mic with REW). It's a little more work, but very possible to get good results.
REW has a IR mode that uses the electric signal as the trigger. Dunno quite what that means but maybe it means something relevant. BTW, sometimes significant time lost inside DSP machines (and different for each input and output) that needs to be accounted for.
Ben
Because of this, I started with attempts at in-room flat direct sound EQ. At the time all my efforts seemed disappointing. My journey over several years led to adjustments to my house curve based on localized power curve measurements (MMM). The MMM listening window repeatability allowed me to reliably chart the impact of even minor changes. It takes extensive work to fine tune the house curve based on listening evaluations. Only a dedicated DIY hobbyist need apply.
It's been several years since I did a direct sound EQ so it is now time to try again. Am I correct in thinking that 1m mic distance and flat response above the Schroeder freq is the correct target? I will use a variable gate the response. In the MF from maybe 300-800 there will still be quite a bit of room impact based on reflections. My TW also has variable SPL to minor positional changes. How do I compensate for these issues in the process? Should I average slightly different mic angles and/or distances? Is there any other suggestions as to the process detail?
Yes, with XLR mic a "loopback timing" is available. This is not possible currently using a USB mic in REW. A solution to the timing issue for those using the USB mic is on the REW wish list.
It doesn't look right. You shouldn't be using exponential averaging. I don't know what happened with the bass extension, maybe you were hitting the noise floor, which tends to be pretty high way down there?
El Greco Fallacy. Equal-loudness contours are not neutral-response targets.
Depends on two things: (1) how peripheral the perceptual mechanism is (couldn't be as extremely peripheral as in the case of astigmatism) and (2) your mental reference.
When you listen to Mahler late at night quietly with the "loudness" compensation switched in (if you use a pre-amp from before 1980) are you hearing the normal balance or is it a poor if satisfying illusion of an illusion?
I think of it like the size:distance invariance principle in vision with your mind choosing what illusory sound stage distance it thinks it is listening to and hence what colour balance (Fletcher-Munson curve) it should be hearing (not unlike the way we assume the distance parameter in seeing the size of objects).
So that's why I think your house curve has to be Fletcher-Munson adjusted but the loudness contour that feels right isn't strictly defined by a mic.
But you could be right.
Ben
Last edited:
Good point that's another measurement I should have done! repeat the mmm measurement with no signal at all. The noise floor in my room is pretty good (at higher frequencies) quite a bit less than 40db (at night the 19db fan in my HTPC sounds quite loud 2.5M away at the listening position), but it is something I should check.
I chose the exponential averaging because of it's slowness (and did the mmm for enough time for the measurement to stabalize. When I tried other averaging when I stopped the mic movements to reach for the stop button in REW the measurement curve changed significantly from what it had just been
Tony.
Last edited:
Hi Hurz, perhaps naive wasn't the correct word. Lacking some of the deeper understanding of acoustics might be more accurate Though a certain level of naivety can lead us down paths that other more experienced people may not even consider. Most times for very good reason, but occaisionally we might discover something unexpected.
I'll see how I go. But don't hold your breath, things move slowly at the wintermute household project wise
Tony.
Though a certain level of naivety can lead us down paths that other more experienced people may not even consider. Most times for very good reason, but occaisionally we might discover something unexpected. I'll see how I go. But don't hold your breath, things move slowly at the wintermute household project wise
Tony.
We've been mulling this over at work recently. I believe that some Fletcher Munson correction is probably needed, but can't imagine it is a perfect correction. The Fletcher Munson curves have always been a measure of how a 1kH tone compares to some other frequency tone. I'm not sure if they were meant as wholesale compensations for music.
The sound stage distance also makes sense to me, meaning if you move away from a sound and it decreases in level, you don't really need bass compensation for that case. The loss of bass from lower listening level is perfectly natural. On the other hand there are probably a lot of cues (direct to reflected level) that come along with the greater observation distance. These cues aren't there if you just turn the volume knob down, so maybe some compensation is needed.
Certainly we have all experienced a system where bass seemed a little light but if we listen 10dB louder it seems about right. The same must occur when we listen to pop music where, say, the mixer was setting balance at a level 20dB higher than we want to play it back at in our room. Some compensation will be needed to get to a satisfying bass balance.
Note that all this refers only to the bass end of a house curve. There is no Fletcher Munson justification for any treble adjustments. Look up ISO 226, the latest curves and subtract the 60 curve from the 80 curve. You will see a straight line bass compensation and little or no treble difference.
The treble aspects of typical house curves have to do with how a flat anechoic, declining power response speaker ends up measuring in the far field of our semi-live room.
David
Last edited:
I normally use time averaging only for phase purposes but if I look at amplitude responses, I would not say there is high frequency loss. The main difference between amplitude averaging and time averaging is
- a not smooth response in mid frequencies for time averaging
- less level in mid/high frequencies in time averaging because in low frequencies, the sync averages are totally coherent so add directly and in high frequencies, they are totally independant.
This is also what jtalden showed.
See here some comparisons between amplitude and time averaged measurements in studios, mixing rooms, living rooms, before any EQ applied... : note that in the first image top left, measuring points are only separated by about 10cm distance
Curious. I said that time averaging of the impulse responses would show a different result and low and behold it does. Which is better or right I never commented on only that they would be different because of the different aspects of the direct (MOSTLY coherent) and reverberant fields (MOSTLY incoherent).
Now Dave can lecture us all once again on why "it's all wrong".
Now Dave can lecture us all once again on why "it's all wrong".
A little bit of revisionist history here?
Only different? Clearly you were claiming that you could isolate the direct sound from the reverberent.
Anyhow, I am happy to "lecture you again" on how it is all wrong.
Time averaging will give a curve wholly dependent on the time coherence of the system from spot to spot. You will have a phase dependent sum, so any phase shift due to observation point will cause cancellation. As a multiway system has multiple acoustic centers, any appreciable change of mic position has to cause some relative phase shift between parts of the frequency band (you understand this, I assume?). A wide enough mic area to start averaging out LF aberrations and room reflections will give an unpredictably erroneous response curve.
Both the curves of jtalden and jlo show midrange variation compared to the MMM average. I assume that they are getting the HF end in phase alignment (1/2 the energy visible in the impulse is in the top Octave, so that is what you Tend to align), so the phase shift ends up at the top of the woofer's range.
Since the only goal here is to get a response curve accurate enough to base equalization on, why pursue something so unpredictable? If you are interested in the direct response then measure it with gating or windowing.
Anyhow, you don't need to argue with me. Show us some curves where you have time averaged over an appreciable area.
Thus endeth the lecture.
David
Only different? Clearly you were claiming that you could isolate the direct sound from the reverberent.
Anyhow, I am happy to "lecture you again" on how it is all wrong.
Time averaging will give a curve wholly dependent on the time coherence of the system from spot to spot. You will have a phase dependent sum, so any phase shift due to observation point will cause cancellation. As a multiway system has multiple acoustic centers, any appreciable change of mic position has to cause some relative phase shift between parts of the frequency band (you understand this, I assume?). A wide enough mic area to start averaging out LF aberrations and room reflections will give an unpredictably erroneous response curve.
Both the curves of jtalden and jlo show midrange variation compared to the MMM average. I assume that they are getting the HF end in phase alignment (1/2 the energy visible in the impulse is in the top Octave, so that is what you Tend to align), so the phase shift ends up at the top of the woofer's range.
Since the only goal here is to get a response curve accurate enough to base equalization on, why pursue something so unpredictable? If you are interested in the direct response then measure it with gating or windowing.
Anyhow, you don't need to argue with me. Show us some curves where you have time averaged over an appreciable area.
Thus endeth the lecture.
David
I still think that you cannot get anything reliable with a measurement at one position, no matter which gating you apply, fixed or frequency dependant.
Measurements with MMM give curves nearer to direct field than to power response. This is contradictory with the fact that we are (or we think we are) beyond the critical distance. I showed some examples in my MMM note. Can somebody provide measurements showing the opposite ? Dave could present a comparison of anechoic measurements with in-room averaged response (MMM or fft averaged multiple positions).
But at the end it doesn't really matter (in fact it matters for most of us) what we are measuring : we should measure what we really hear. I'm sure that on this point, even Earl and Dave agree...
So I have following proposal. We know that :
- absolute quality is hard to evaluate
- comparison listening is difficult because it is very subjective to tell what is best, there are so many criterias and subjectivity is so unreliable
- comparison listening is easier if you just have to tell "same" or "not the same"
So I thought of following test setup, in mono :
- one loudspeaker is reference A (in the sense of an "anchor", not in the sense of highest quality)
- another loudspeaker is B, different brand or model
Place both loudspeaker quite near from another.
Measure both loudspeakers with different methods and prepare two comparisons :
- Equalize only loudspeaker B with various methods to get the same response as loudspeaker A uncorrected.
- Equalize both loudspeaker A and B with various methods to same target curve.
Now in both cases, compare A and B : the best measurement+EQ method would be the one for which A and B now sound the more similar, this could be checked with an ABX test. I know this criteria is very difficult (and maybe impossible for loudspeakers) but let's consider it is feasible.
I'm preparing such a test, here is my setup :
I choose two loudspeakers I had on hand, the reference is the B2030a because its bass extension is less than p360 (would be more difficult to EQ the B2030a to have the bass of the p360 with risks of distortion,...)
Any comments on the validity, problems or improvements of this test ?
Well, I've been EQing speaker A to sound like speaker B in room X since I was a little kid (don't really know why), and I still do it fairly often, and now I have test equipment so in more recent times I could see what I was doing.
I've never done it with the goal of coming up with an EQ recipe, but guessing/estimating from memory of doing it a bunch of different times with all sorts of different speakers, I think it has always turned out pretty much the same. For the closest subjective in-room match, you EQ on-listener-axis anechoic response to match, then EQ power response to match up to some lower midrange level, then sort of split the difference between power response and on-axis anechoic response from there on up.
It's entirely feasible to get a close subjective match with even pretty different speakers as long as they aren't too terribly behaved and you're not running out of output. "Close" as in "yeah that sounds like the same frequency balance", not "I can't tell the difference at all", so I don't think ABX is reasonable; you'll just get a passed ABX for all attempts unless you're using very similar speakers to start with. You need to ABC, "which is closer to A?", ACB "which is closer to A?", etc.
I know there are no amazing insights there, but I guess my point is that with a human that has some basic knowledge of how speakers work at the reins, you can use any number of measurement methods to do pretty much the same thing and there's no real problem. It's only when you try to come up with a universal step-by-step method that you have a problem. Is that now the goal of this thread? It seems like the threads discussing Acourate, Dirac, and so on have been down that road pretty thoroughly :shrug:. I mean, the book isn't closed, but this feels kind of like starting it over?
I've never done it with the goal of coming up with an EQ recipe, but guessing/estimating from memory of doing it a bunch of different times with all sorts of different speakers, I think it has always turned out pretty much the same. For the closest subjective in-room match, you EQ on-listener-axis anechoic response to match, then EQ power response to match up to some lower midrange level, then sort of split the difference between power response and on-axis anechoic response from there on up.
It's entirely feasible to get a close subjective match with even pretty different speakers as long as they aren't too terribly behaved and you're not running out of output. "Close" as in "yeah that sounds like the same frequency balance", not "I can't tell the difference at all", so I don't think ABX is reasonable; you'll just get a passed ABX for all attempts unless you're using very similar speakers to start with. You need to ABC, "which is closer to A?", ACB "which is closer to A?", etc.
I know there are no amazing insights there, but I guess my point is that with a human that has some basic knowledge of how speakers work at the reins, you can use any number of measurement methods to do pretty much the same thing and there's no real problem. It's only when you try to come up with a universal step-by-step method that you have a problem. Is that now the goal of this thread? It seems like the threads discussing Acourate, Dirac, and so on have been down that road pretty thoroughly :shrug:. I mean, the book isn't closed, but this feels kind of like starting it over?
I actually have a measurement that I took a number of years ago. This is specifically the room gain of 2 rooms and 2 types of speakers (all 4 combinations averaged) as I measured at PSB a few years back. We were looking into using Audessy and I was searching for a sensible room curve.
What I did was to measure a speaker in a listening position at 1/2 meter from the system baffle at a spot between woofer and tweeter. From that position I moved in a straight line towards the listening position while taking more measurements at 1, 2 and 3 or 4 meters (depending on the room size). All the later curves are divided by the 1/2 meter curve, leaving only the difference curve. Assuming the 1/2m curve is largely anechoic, the subsequent curves show the difference due to moving away in the room. (speaker near field normalized out)
I've averaged the 4 curves together. You can think of them as generalized room gain curves with some compensation for the average power response of the 2 systems.
You can see that the movement from near to far resulted in about a 2dB shelving down of response from the low treble, with about 5dB total loss at extreme HF frequencies. There is also a significant LF room gain that may or may not be desirable.
These only apply to the particular rooms, but I think they are typical of the response change from near field to far field.
The green curve is the actual calculated average and the blue ink drawing is my approximation to it.
David
Attachments
I said enhance - "enhanced" is not "isolate".
You said that the HF would drop, they didn't. The fact that the synchronous averaging is different that the non-synch means that something different is happening and nothing in what you said proves that it is not an "enhancement" of the direct field.
Yes, strange, strange, strange.
Does an outdoor band sound "wrong" as you walk away or does it sound exactly like an outdoor band at a distance? Does late-night loudness-compensated Mahler sound very bassy when you think about it but also just right when listening to it?
One thing is hard to argue with: the underlying rationale of the Fletcher-Munson curves is unassailably correct, in vitro. (The 1kHz is an arbitrary reference and curves can be created step-by-step as well.)
For sure, playing music at the right level is very important but I sure have no coherent explanation why.
About Fletcher-Munson treble, yes there's definitely an El Greco fallacy going on since the curves are parallel. What's right at 40 is right at 80 and what's right live ought to be right when reproduced*. On the other hand, we are back to the unassailable logic that says treble needs to be louder to sound the same as mid-range, in vitro.
Perhaps there is a connection between best playback level and hitting a good Fletcher-Munson balance.
Ben
* "taste" confuses the issue... some of us like extra treble to add (unnatural) snap to percussion instruments and everybody likes more bass than you'd hear at Carnegie Hall (or even that great bassy hall in Philadelphia)
Last edited:
Certainly early reflections cause disruptive interference and further down the time line can enhance if specturally balanced. Very late delayed reflections obscure speech. But what is the time line between very early disruptive and further delayed enhanced reflections specifically? Define "enhanced" for technical clarity please.
Nice to see data of that sort.
But for each distance (which is also a time difference) shouldn't there be a correction for room reflections present at that moment in time and within the measurement window? And would that make your final curve all but flat?
(Just my personal soap-box, but I wish "room gain" would be used only for the mythical boost, as Linkwitz points out, in bass below lowest mode in tighly sealed concrete bunkers and well-sealed cars. Otherwise, lots of acoustic factors in rooms favouring bass but not properly called "room gain.")
Ben
Last edited:
- Status
- This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.