An area average is great if you want to satisfy several people across a range of seating spots. If you want the best possible response at one position ("the money seat" they call it in home theater) then an exact EQ of that spot will give the flattest response there. The wider your measurement area the more it becomes a compromise.
I base this on lots of years with both manual and automatic EQ systems. If you have automatic EQ with a single mic position you get a very precise inversion of the room curve. Use that system with a multiseat measurement average and the average is, of course, flat but each individual seat deviates. (Seat to seat variation is never effected by EQ)
As to averaging out reflections, you need to decide if the reflections are audible. In my experience, cabinet edge reflections, floor bounces and wall bounces are generally audible if they are significant in the measurements. If your seating position is well constrained it is worthwhile to EQ them. If your seating area is not constrained then you shouldn't.
David
I base this on lots of years with both manual and automatic EQ systems. If you have automatic EQ with a single mic position you get a very precise inversion of the room curve. Use that system with a multiseat measurement average and the average is, of course, flat but each individual seat deviates. (Seat to seat variation is never effected by EQ)
As to averaging out reflections, you need to decide if the reflections are audible. In my experience, cabinet edge reflections, floor bounces and wall bounces are generally audible if they are significant in the measurements. If your seating position is well constrained it is worthwhile to EQ them. If your seating area is not constrained then you shouldn't.
David
In my note about the MMM method, I presented many measurements done at only 10cm distance. The differences are quite important. Which of those measurement will you base your EQ on ? Should we EQ for the left or the right ear ? I'm quite sure that for most of us, just moving 10cm, the perceived sound of our system does not change.
I have the feeling that one measurement at one place is not representative of what we hear. Hopefully some disagree so we can have nice discussions and allways learn on the very interesting relation between measurement and perception.
I think your curves show some variation in the crossover region of your system and you are advocating using a blended (by moving microphone method) EQ of the average curve.
The question always gets back to "do we hear these variations?" If we can hear the variations then an average correction is just that, an average. A lot of the "fuzz" on your curves is narrower than a critical band and probably worth ignoring, but the midrange variation is wider and likely audible. Certainly if it is variation in crossover blending due to position, that is something that I have always been able to hear.
As to one measurement at one place, it would make no sense to think that the ear can in any way sense response at positions other than its current location. Averaging across space and then sitting in a fixed location doesn't seem like it will be beneficial. In the case of EQing a theater with many seats then a wide area average makes sense and is standard practice. If you are a solo listener and not going to move, then you are better served by doing a more localized measurement and EQ. (Yes, for one ear position and the other if the measurement variations warrant it.)
I have always thought of room response as something that should be looked at versus a listener location "probability cloud". That is, measure at the likely listener axis and around it to a distance that just covers the probable listener position variation. Ideally you would weight each curve by the likelihood of a listener being at the position. When the probability cloud is wide then a wide spatial average is required. If listener position can be constrained then a wide area average will only degrade the results.
David S.
The question always gets back to "do we hear these variations?" If we can hear the variations then an average correction is just that, an average. A lot of the "fuzz" on your curves is narrower than a critical band and probably worth ignoring, but the midrange variation is wider and likely audible. Certainly if it is variation in crossover blending due to position, that is something that I have always been able to hear.
As to one measurement at one place, it would make no sense to think that the ear can in any way sense response at positions other than its current location. Averaging across space and then sitting in a fixed location doesn't seem like it will be beneficial. In the case of EQing a theater with many seats then a wide area average makes sense and is standard practice. If you are a solo listener and not going to move, then you are better served by doing a more localized measurement and EQ. (Yes, for one ear position and the other if the measurement variations warrant it.)
I have always thought of room response as something that should be looked at versus a listener location "probability cloud". That is, measure at the likely listener axis and around it to a distance that just covers the probable listener position variation. Ideally you would weight each curve by the likelihood of a listener being at the position. When the probability cloud is wide then a wide spatial average is required. If listener position can be constrained then a wide area average will only degrade the results.
David S.
The problem (for me) with one location is that my ears and head are not 1/4" across. My microphones are. What the microphone measures at one spot or another often does not jibe with what I subjectively hear. No surprise, I don't hear a spot 0.25" across.
Doing averaging over multiple locations does help reconcile measured vs perceived EQ. The MMM seems fast, repeatable and gets close to what I hear within the space where I sit. To me the difference is large between MMM and static EQ. An EQ based on MMM is much more pleasant to listen to. It has actually pointed out flaws that I could hear or sense before, but had difficulty measuring. Certainly it is not perfect, but for me it's been a big step forward.
Doing averaging over multiple locations does help reconcile measured vs perceived EQ. The MMM seems fast, repeatable and gets close to what I hear within the space where I sit. To me the difference is large between MMM and static EQ. An EQ based on MMM is much more pleasant to listen to. It has actually pointed out flaws that I could hear or sense before, but had difficulty measuring. Certainly it is not perfect, but for me it's been a big step forward.
The problem with single point measurements is statistical significance, just as jlo says (not to mention Schroeder). A single point is simply not statistically significant unless some averaging is used. You can average across frequency, but that's not a good idea, or you can average in space, which is the correct way to do things.
When we wrote our paper back in the 80's we showed a +- 6 dB variation where there was a virtual certainty (90% chance) of some listeners ears being located.
And then there is temperature and humidity. I show in my book how typical weather variations will change the absolute transfer functions to a single point by several dB. Spatially averaged measurements never show any change.
Spatial averaging is simply an essential element to valid EQ.
Although I have to say at this point that to me this whole discussion misses the critical point. Do we want to EQ the reverberant field if the direct field is flat and smooth? I say not. Only an anechoic response of the loudspeaker can tell you if the direct field is flat or not. You can't determine that in a room - unless it is very big, certainly not any home listening room.
When we wrote our paper back in the 80's we showed a +- 6 dB variation where there was a virtual certainty (90% chance) of some listeners ears being located.
And then there is temperature and humidity. I show in my book how typical weather variations will change the absolute transfer functions to a single point by several dB. Spatially averaged measurements never show any change.
Spatial averaging is simply an essential element to valid EQ.
Although I have to say at this point that to me this whole discussion misses the critical point. Do we want to EQ the reverberant field if the direct field is flat and smooth? I say not. Only an anechoic response of the loudspeaker can tell you if the direct field is flat or not. You can't determine that in a room - unless it is very big, certainly not any home listening room.
Using REW, I took 6 individual measurements of my right speaker across a 6' x 2' grid that covers my listening area (couch) some 9' away from the speaker.
My understanding is that 1/6 octave smoothing represents the ears critical bandwidth (more or less) Music and the Human Ear So, this is the smoothing applied to the charts below. I am happy to supply different smoothing as well as the impulse and/or step responses...
The REW sweeps were from 10 Hz to 24 kHz with a 500ms (default) REW window.
These measurements are from a 3 way floor stander with 15" woofer in a sealed cab, old school midrange compression driver/horn combo, and BMS 4540 on a QSC waveguide. I time aligned the drivers using digital XO points of 500 Hz and 5 kHz and eq'd with Acourate (which uses Frequency Dependent Windowing (FDW) and a proprietary psychoacoustic filter. I only used one measurement position to design the FIR filters. For verification, I was able to send REW's sweep signal through JRiver's Convolution engine, with the Acourate designed FIR filters engaged, and took the following 6 measurements across a 6' x 2' grid area - one sweep at a time.
My understanding is that 1/6 octave smoothing represents the ears critical bandwidth (more or less) Music and the Human Ear So, this is the smoothing applied to the charts below. I am happy to supply different smoothing as well as the impulse and/or step responses...
The REW sweeps were from 10 Hz to 24 kHz with a 500ms (default) REW window.
These measurements are from a 3 way floor stander with 15" woofer in a sealed cab, old school midrange compression driver/horn combo, and BMS 4540 on a QSC waveguide. I time aligned the drivers using digital XO points of 500 Hz and 5 kHz and eq'd with Acourate (which uses Frequency Dependent Windowing (FDW) and a proprietary psychoacoustic filter. I only used one measurement position to design the FIR filters. For verification, I was able to send REW's sweep signal through JRiver's Convolution engine, with the Acourate designed FIR filters engaged, and took the following 6 measurements across a 6' x 2' grid area - one sweep at a time.
That is a strong statement that needs corroboration. Clearly frequency smoothing to approximate critical bandwidth makes sense. Spatial averaging will smear the variations due to listener position. If you can't fix the offending variation or if you can't fix listener location then it may be your best course, that does not mean that the averaged out aberations were inaudible.
I would agree with the first part of that. But the example shows a positional midrange variation that looks like a crossover region blending problem, a direct field effect. The author should not claim that is inaudible.
As to the direct sound in a live room, we measure that all the time with windowing approaches. It works quite well for midrange frequencies and up. For lower frequencies the steady state response is more appropriate, but we can do live room measurements with windows that well match perception (read Kates or Salmi regarding that).
David S.
Serious!!? How could dB level claims be done by ear? Do I do anything by ear? It was a ridiculous question.
The example I presented at the beginning of my note shows response irregularities far from crossover point (here 530Hz). The measurements are done at 7m and separated by a 10cm distance, this is such a small angle that variations cannot be due to crossover blending.
If you look at the above example given by Mitchba, the variations are also far from crossover frequencies.
I do consider that those frequency irregularities are coming from reflections and are not a direct field effect.
I have also used a time windowing approach, frequency dependant or not, but was never completely satisfied :
you do a nice correction but when you re-measure at a position not exactly the same, you see variations on the curves. As I stated in my MMM note, even with windowing and smoothing, those responses, despite very near positions, show real differences.
As shown at the end of the MMM presentation, the MMM measured response in mid frequencies and up, is very near from anechoic measurements, between what Olive calls listening window (LW) and early reflected curve (ER), (see : A Multiple Regression Model for Predicting Loudspeaker Preference Using Objective Measurements: Part II - Development of the Model)
And as he concluded in part 1 of the paper : "The perceived spectral balance strongly correlates with the loudspeaker’s measured listening window, giving credence to the importance of the direct sound in timbre perception"
By chance, MMM measures something not far from LW "listening window" !
That's why it would be interesting if others could compare MMM to real anechoic measurements.
Such is life.
Many speakers do vary across their listening window. If it isn't a crossover effect then a horn pattern effect or? When the variation is fairly broad and of more than a few dB, then you would expect it to be audible and for the sound to change as you move through those positions. Doing a spatial average may make the aberration go away in your measurement and suggest a good average EQ, but that does not prove that it was right to ignore the aberrations. The midrange variations shown in your curves would be audible.
When those variations are due to direct field, sure it would be very audible.
But when those variations are due to reflections coming from another direction than the loudspeakers, I think that our auditory system has some ressources to minimize their influence and concentrate on direct field.
People (Dave?) need to read Schroeder's paper. The professor assumed that the source was perfect in every way and yet there were still variations in the field response. It is not a loudspeaker effect. The variations are not really audible as they are too narrow, but they do make specifying EQ more complicated because you don't really have a real stationary measure, but a more random one. I completely agree with jlo on all of this and I think that he has a very solid understanding of the situation.
You said:
I see no decibel claims there. Since you had mentioned psycho-acoustics previously, I thought you might be talking about subjective perception of tonal balance. Seems I was wrong about that.
FWIW, there are pros who are quite good at decibel levels by ear. I've measured the claims. They fit. Just because some of us can't do it, doesn't mean no one can. (I'm not very good at it.)
Not that I know of. Probably not even listening to the TV or telephone. But why are you talking about you?
Always the gentleman.
Always the gentleman.
I treat others as I am treated. Your question sounded provocative, we know that you do that. Maybe not, then my mistake.
I primarily do all testing in mono, single speaker setup. One of the aspects of my livingroom (that doubles as a ******* acoustics lab) is reflections off of my Samsung 46" lcd tv. The response is quite even off axis, but I can clearly pick out the reflected midrange band. Reflection from tv is about 6ms delayed coming from the left from my listening position, about 5 meters away from the source. Speaker placed roughly mid wall and against it (part of the design) offset to the right on the far wall in front of me. Right wall delay is even greater and is somewhat noticeable but not nearly as pronounced.
My findings from 30 years or so ago and this solidify in my mind that we can reject most off axis amplitude changes if level is low enough. Almost as if the rate of room excitation has an effect on this audibility. Tho I do tend to think it is something other than this, another psychoacoustical property perhaps. Listening in stereo would clearly mask what is heard otherwise.
I don't feel the need to debate you on every thread but you once again misconstrue or ignore what I have said.
The author shows some midrange variations of over an Octave wide and +-4dB in magnitude. These are unrelated to any room related Schroeder type variations. They are wider than a critical band, and very likely audible.
If it is not a loudspeaker effect it is certainly a nearby reflection effect (quefrency is low) and not one that should be dismissed as inaudible. (by you or the author).
It is very true that a high resolution unsmoothed room response is a very messy thing with much information showing discrepancies that we just can't hear. Divining the audible from the merely visible is a crucial issue and there are still things we don't know about perception. (Early sound vs. late, effect of direction of arrival, resonance vs. reflection, etc.)
The point I have made and will stick with is that, just because a measurement approach glosses over certain aberrations and gives a smooth curve easier to use for equalization, is no proof that it was proper to do so!
The author needs to determine the cause of those aberrations and justify his belief that they can safely be ignored.
David S
- Status
- This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.