Wouldn't it be great to have a secret guide to the web names of distinguished people. Otherwise, we have to recognize them by the aptness of their posts, as in the helpful post from "David" of Framingham.
Dunno if you looked at my Giotto post and paradox of the creation of illusion of a scene from a piece of shiny flat canvas.
Recently, perhaps re-inspired by two Griesinger videos, I've been thinking about the perceptual cues of listening to music and as provided (or more often, absent) in reproduced music. These cues and their relative "ecological validity" or weighting have been amply studied for vision (see Gestalt psychologists and Adelbert Ames) but very little for sound and then mostly conjecturally.
I suspect that good systems (or halls) allow the listener to "stream" the illusory (or real) orchestra as a coherent "perceptual event*" because they provide cues that allow the listener to complete a good functioning mental construction of that event.
Whether with speakers or live musicians, direct sound is better at providing those cues and better direct sound does it better.
Ben
*sound perceptions take place in time and so are properly called events, in contrast to visual objects which take place in space
Dunno if you looked at my Giotto post and paradox of the creation of illusion of a scene from a piece of shiny flat canvas.
Recently, perhaps re-inspired by two Griesinger videos, I've been thinking about the perceptual cues of listening to music and as provided (or more often, absent) in reproduced music. These cues and their relative "ecological validity" or weighting have been amply studied for vision (see Gestalt psychologists and Adelbert Ames) but very little for sound and then mostly conjecturally.
I suspect that good systems (or halls) allow the listener to "stream" the illusory (or real) orchestra as a coherent "perceptual event*" because they provide cues that allow the listener to complete a good functioning mental construction of that event.
Whether with speakers or live musicians, direct sound is better at providing those cues and better direct sound does it better.
Ben
*sound perceptions take place in time and so are properly called events, in contrast to visual objects which take place in space
Last edited:
I'm not sure which non-flat curve that is unless you mean what they were shooting for in their comparison of automatic DSP EQ systems.
I did sit through a paper by Olive on headphones (Harman has published 5 headphone papers in the last year or so).
It seems the headphone people are as uncertain about what exact target to use ase we are. He claimed the industry was split between those shooting for an (at the eardrum) response as would match one heard in a flat diffuse field, and conversly in a flat anechoic field.
He did a blind study where both those approaches were ranked lower than headphones EQed to match the in- room response of the big JBL (K2?) monitors.
I especially enjoyed the paper since his data showed the least favorite headphones in a comparison, by a wide margin, were by BEATS!
David
Hi jlo
I have a real problem with this claim. Maybe we aren't using the same definitions of terms or something like that. To me there are two different "direct" definitions. The most common is the direct field close to the source such that the sound level from the source dominates the total sound level in that region. This is opposed to the reverberant field where the sound level is dominated by the reflected/reverberant sound levels. The transition between these two is called the critical distance.
To me, there is another definition of the direct sound, which is counter to that of the direct field. And that is the sound that arrives at the listener prior to any reflections. This sound exists even beyond the critical distance. Since it lasts for only a small time it is not really a "field". That is why I distinguish the direct sound from the direct field. The direct sound is measured as the anechoic response of the source along the listening axis. It is independent of the room. The direct field depends on the room.
Your MMM technique cannot in any way detect the direct sound since it is steady state. It measures a combination of the direct field and the reverberant field depending on where one is relative to the critical distance. You seem to be implying in your statement above that the average of the moving measurements is such that the reverberant field portion is somehow averaged out or at least lowered. I would make two comments about this claim, if in fact that is what you are saying; 1) it is a hypothesis given without any real support and; 2) it contradicts the original Schroeder paper which clearly shows that the reverberant field does not average to zero, indeed in converges on the mean power response which is not at all zero. This is precisely what Schroeder's paper is about. The statistical fluctuations in the sound field do not average down as more and more points are taken, but they in fact average to their mean, some frequencies dropping in level and others rising.
If I misunderstand your claim then please correct me, but if I do understand your claim correctly then please highlight why it is that you believe that the reverberant field will average down below the mean sound power response of the room.
Last edited:
Thanks Dave - this was most enlightening. Especially since this is what I have been saying in this thread from the beginning. I'd prefer we used the term "listener axis response" as opposed to "axial response" since the two need not be the same, but that's a small point.
That's also how I see it. I think that MMM is a quick way to do spatial averaged in-room measurements that are reliable enough to permit precise EQuing. But with loudspeakers having a not smooth directivity, as MMM shows more of the direct response, your EQ will be more a correction of direct sound than of power response.
In most in-room measurements, we are above critical distance, even at the highest frequencies.
You are right, I should use a better terminology : the reverberant field is not averaged out but it is smoothed. The reverberant response at one point has a chaotic response, spatial averaging simply reduces this chaos.
You allways end up with a response that is a blend between direct sound and reverberant field. But you see in the real measurements, that you get much more from direct sound (speaker response between listening window LW and early reflections ER , as per Olive/toole terminology) and less from reverberant field (speaker sound power response plus room response).
Earl, I think you take part to the SMPTE B-chain group and a draft TC25css-B was just published a few days ago, at page 82, point c) : "This indicates that the direct field strongly dominates the measurements, and that reverberant build-up contributes only a small amount of sound to the total energy at these frequencies, even deep into the rooms." This is based on measurements in theaters but for what I've seen, it is the same in many living rooms.
I also remember the JBL Sound System design, at last page, last picture : https://www.jblpro.com/pub/manuals/pssdm_1.pdf. In our usual "not so live" rooms, the observed attenuation at a distance is higher than the calculated theorical distance decrease.
Finally, we are measuring in an acoustic field far from really diffuse.
Moreover, when we measure near listening place, we stay more or less within loudspeaker axis, this is another point that gives more weight to direct sound in the measurement.
Does all this mean that the reverberant field is in fact lower than supposed compared to direct sound ?
My claims are :
- spatial averaging (with some constraints) gives a good approximation of direct sound (between LW and ER, see above).
- MMM is a simple and quick tool to do spatial averaging
The second point is easy to check and I have already shown comparisons of MMM and multipoints measurements.
On the first claim, main constraints are :
- measure within a given solid angle around listener axis response
- measure at enough independant points
- the claim is only valid above transition frequency.
You end up with a measured response that is nearer to direct sound than to power response.
At the end of my MM note, there is a comparison of the Harman measurements of Infinity P360 and same speaker in my living room measured with MMM.
Other example, here is a comparison of anechoic measurement and spatial averaging, not so far ?
Re-reading my post, I think I should have used direct field instead of direct sound at some places !
I want to add something : as I explained in the MMM note, the most important claim is that a unique point measurement has poor relation to our perception. The perceived timbre is the same when you move your head but measurements varies a lot. Averaging is a way to find common factors to different positions. For sure there is a question of what is the proportion of direct and reverberant field in the measurement, but at the end the problem is more : does this way of in-room averaging correspond to what is really heard ?
I want to add something : as I explained in the MMM note, the most important claim is that a unique point measurement has poor relation to our perception. The perceived timbre is the same when you move your head but measurements varies a lot. Averaging is a way to find common factors to different positions. For sure there is a question of what is the proportion of direct and reverberant field in the measurement, but at the end the problem is more : does this way of in-room averaging correspond to what is really heard ?
Not by a long shot, even if it is a helpful method of removing room-mode sound.
1. Only a waterfall display can present (some, most, or all) of the information needed.
2. Only when we have good understanding of how direct, reflected, and room-mode sound and their time courses relate to hearing perception.
Ben
I don't think I can agree that the MMM technique focuses in on the direct field. At any point in space the room's direct to reflected ratio is already in effect and moving the mike around (at a fixed distance from the source) doesn't change that relationship. If anything you are beginning to sample the direct response over a range of angles and, in the extreme, starting to see more of a power sampled view.
Speaker power response is always a factor in any room curve at a typical distance. I believe we tend to be a little beyond (is that your "above"?) the critical distance for mid frequencies in most situations. This means the room response is somewhat more weighted to the power response than to the direct response.
This is substantiated in the Toole measurements that I saw last week, with power response wiggles clearly seen in measurements at the listener position.
If we are all agreeing that the direct response is important and that EQ of the direct response is the best way to go, then why not just measure the direct response? Especially if you know where you will be sitting, why not just place a microphone and do a gated measurement? I like Holm impulse which gives an option for a variable bandwidth window with time window length inversely proportional to frequency, just as the doctor ordered.
MMM will give a room curve with results averaged over the area encompassed. If you want a room curve and will figure out a suitable room response target, then that is fine. If you think there is uncertainty in where you will sit or you must cater to a few listeners, then that is fine. But lets not fool ourselves into thinking we are measuring the direct response when we aren't.
In the end this (MMM) will always be a steady state measurement with no discrimination between early arriving and late arriving sound. As such it sidesteps the question of how we perceive frequency response and forces us to guess at what target curve will sound good in this room with this speaker (of unknown directivity index curve).
As to diffusion, all rooms are well diffuse above a certain frequency. That was the purpose of Schroeder, to define frequencies below which rooms are no longer diffuse. Room acoustics are highly scalable. If a concert hall is highly diffuse, then a 1/10th scale model is similarly diffuse for wavelengths scaled by the same factor. It is not room size but size relative to wavelengths under consideration, that matters.
Regards,
David
Just because it doesn't work (well I think it doesn't work....)
Here is a frequency dependant gating, done with HOLM, of some of the measurements that I showed in my MMM note. The gating corresponds to a complex smoothing with N=6 which is quite short.
Other measurements separated by 40cm in a mixing room (X curve) :
I can tell that those curves are very far from the anechoic measurements.
Would you rely on those responses to EQ ?
Last edited:
No, I would not use that data for EQ. You have too many very early reflections.
I would EQ the anechoic response of the loudspeaker along the listening axis and then not change it in-situ except for frequencies < something like 200-400 Hz. where the modes interfere with the speaker, i.e. the room and the speaker couple to make a single response. At frequencies higher than this the room and the speakers do not couple.
I would EQ the anechoic response of the loudspeaker along the listening axis and then not change it in-situ except for frequencies < something like 200-400 Hz. where the modes interfere with the speaker, i.e. the room and the speaker couple to make a single response. At frequencies higher than this the room and the speakers do not couple.
Absolutely.
You are both missing the point. The objective is not an easy anechoic curve or a smooth and pretty in-room curve. We must equalize what is perceived by ear.
Your curves show some very significant early reflection, such as a floor or console bounce. Surely those are audible?
Kates did a paper where he used an auditory model to see how we perceived frequency response in a live room. He showed that a time window of varying length, long at low frequencies and short at high frequencies was the answer. In his tests the model encompassed only the direct sound at high frequencies but was long enough to capture floor bounce, similar to your aberations, at mid frequencies.
Bech did simulations of all the reflections of a typical room and found, similarly, that the reflections likely to be audible were the floor bounce and back wall bounce.
It may not be easy to EQ and you may not want to fill in the dips totaly with huge amounts of boost, but those aberrations are real and should be dealt with.
If we look to optimize the anechoic response above 2k, the anechoic plus first bounce from 200 to 2k, and the steady state curve from 200 down, then we won't be far off.
David
JBL/Toole typically EQ the "listening window" response, that is averaged anechoic (gated) response in the 30°/10° window.
That seems to be a reasonable approach as it minimizes diffraction effects but still focus on the direct field only.
Here is an example for the JBL M2 loudspeaker.
This one even has a rising on-axis UHF response.
Attachments
David - your "missing the point"
Audible, sure but should they be corrected with EQ as opposed to good loudspeaker design and room setup? No. We must differentiate between the right way to do something and Band-Aid methods. The EQ of a poor in-situ response is a Band-Aid. It may improve things, but sometimes they get worse (as Olive has shown), clearly they are not optimal.
The rational for a narrowing time window with frequency can be seen from critical band theory where the impulse responses of the Gamma-tone filters get longer and longer at ever lower frequencies. But the narrowing is not linear in frequency as it is in Holm. It is far more complicated than that. Using critical band smoothing, as proposed by Farina (as I mentioned before) is a good way to approximate this effect. Better would be to use Gamma-tone impulse responses to decompose the rooms impulse response thereby capturing both the ERB bandwidth and the finite time window aspects of human hearing simultaneously. I am writing the code to do this as we speak.
While Kates and Bech have shown what might be audible, there is no evidence, that I am aware of, that correcting the kinds of problems that we see in the measurements above is optimal or even effective. It is better to correct acoustic problem in the acoustics domain, not in the electrical one.
First, I don't see why the JBL approach "minimizes diffraction effects".
I don't have a strong objection to the JBL approach, but I have some issues with how it was derived. They went out to customers and measured the customers in-situ very early reflections. Then they used this data to create their "listening window". The question that I have is: Why are we to assume that a finite set of their customers have the best possible room setups? If they were all equally bad setups then we will have standardized on a measurement that incorporates these bad aspects.
If we look at first image of post 191, I simply cannot understand which curve you would choose to base your EQ on : at 900Hz, there is a difference of about 16dB between the red and the green curve ! The measuring points are only about 20cm apart and I can assure you that the timbre is not changing a lot when you move just 20cm.
jlo - you make some good points. Clearly some averaging must be used. Spatial, frequency, windowing?
My only problem with what you have been saying is that I do not agree with the claim that MMM gives us either the direct response or the direct field. To me it gives us the reverberant field (as long as we are > critical distance.) Do we want to EQ the reverberant field? Sounds like that is a current topic of discussion across a wide array of people.
My only problem with what you have been saying is that I do not agree with the claim that MMM gives us either the direct response or the direct field. To me it gives us the reverberant field (as long as we are > critical distance.) Do we want to EQ the reverberant field? Sounds like that is a current topic of discussion across a wide array of people.
Toole, Olive,... showed that
- PIR predicted in-room calculated response
- ER Early reflections
- real in-room measurements at listener place
The real MMM difference, is that, instead of giving precisely the awaited ER or PIR equivalent curve, MMM gives a response between ER and LW listening window. As if the ratio between direct/reverberant is, in many cases, a bit more weighted toward direct field than theory would indicate. I don't know why.
But if we read Olive AES convention paper 6190 Predicting Loudspeaker Preference Using Objective Measurements, figure 7 :
Do you also see a kind of a slope above 500Hz ? Could it be that Harman's measurements are also showing a bit more direct field than predicted ?
I think that most of us don't want to EQ reverberant field because we think that direct field is the most audible part, at mid and high frequencies.
jlo
When you try and talk about the MMM and the direct field, Early Reflections or the direct sound, I fail to follow your arguments.
How could the PIR and the Early Reflections "all generally very near one from another". They are not comparable things. I agree that the PIR and the "in-room measurements at listener place" will be close, they are basically the same thing.
Toole says that the In-room response is highly correlated with the speakers power response (makes sense.) The power response is not always correlated with either the direct sound or the listening window average. (In a "good" loudspeaker it should be.) So there is no guarantee that the in-room response will be correlated with the listening window in an arbitrary loudspeaker.
Sounds like a bit of hand waving to me (but then I don't really follow why you say that.)
Agreed that MMM is an ideal way to measure the in-room response, but that this measure is not heavily weighted towards the reverberant field is something that would need to be proven to me.
Where your claim might be valid, but this remains to be shown, is that the direct field CAN fall slower than the predicted 6 dB per DD. This will tend to shift the ratio of the direct to reverberant field contribution above the critical distance. But how common is this change in falloff? I have seen it measured in some cases, but it is not always true. When does this occur and when doesn't it? These questions all remain even if the classic falloff can justify a greater measure of the direct field by MMM in the reverberant field.
When you try and talk about the MMM and the direct field, Early Reflections or the direct sound, I fail to follow your arguments.
How could the PIR and the Early Reflections "all generally very near one from another". They are not comparable things. I agree that the PIR and the "in-room measurements at listener place" will be close, they are basically the same thing.
Toole says that the In-room response is highly correlated with the speakers power response (makes sense.) The power response is not always correlated with either the direct sound or the listening window average. (In a "good" loudspeaker it should be.) So there is no guarantee that the in-room response will be correlated with the listening window in an arbitrary loudspeaker.
Sounds like a bit of hand waving to me (but then I don't really follow why you say that.)
Agreed that MMM is an ideal way to measure the in-room response, but that this measure is not heavily weighted towards the reverberant field is something that would need to be proven to me.
Where your claim might be valid, but this remains to be shown, is that the direct field CAN fall slower than the predicted 6 dB per DD. This will tend to shift the ratio of the direct to reverberant field contribution above the critical distance. But how common is this change in falloff? I have seen it measured in some cases, but it is not always true. When does this occur and when doesn't it? These questions all remain even if the classic falloff can justify a greater measure of the direct field by MMM in the reverberant field.
Last edited:
- Status
- This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.