MMM approach and a new calibration app (magic beans)

[DWPress]

I can see the logic here and a MMM use I hadn't thought of before. With using active crossovers with 3 way speakers + 2 subs generated via software I have done near field sweeps of each individual driver and made PEQ corrections. I figured if the driver measured well in isolation at the baffle then the sum of all drivers on the baffle would also improve. Doing it via MMM a better idea as RTA info is all that is needed to get eq and spl levels and cover a greater area of driver radiation.

This method does a similar thing with multi ch speakers and the room instead of the baffle and will give you a theoretical room curve. This is where I see the real value. None of the Harman, Katz or other commonly shared elevated bass curves have ever sounded good in my room when tested but it would be great to have something more to start with than a gradual decrease in dB up to 20k.

Have you tested it with 2.x channel systems at all?

 

[joentell]

I can see the logic here and a MMM use I hadn't thought of before. With using active crossovers with 3 way speakers + 2 subs generated via software I have done near field sweeps of each individual driver and made PEQ corrections. I figured if the driver measured well in isolation at the baffle then the sum of all drivers on the baffle would also improve. Doing it via MMM a better idea as RTA info is all that is needed to get eq and spl levels and cover a greater area of driver radiation.

This method does a similar thing with multi ch speakers and the room instead of the baffle and will give you a theoretical room curve. This is where I see the real value. None of the Harman, Katz or other commonly shared elevated bass curves have ever sounded good in my room when tested but it would be great to have something more to start with than a gradual decrease in dB up to 20k.

Have you tested it with 2.x channel systems at all?

It seems like you get it.

We have tested it numerous times on 2.x systems and even in mono.

There's no "magic" in the method. Anyone can try it for themselves in REW. Take a NF MMM of a speaker, and MLP MMM and use trace arithmetic to divide one by the other (I can't remember which to divide over what at the moment, but try both and see which one provides a downward sloping curve.) That should be your room response for that speaker at that single listening position. How does that bass rise look compared to the other curves you mentioned?

Most of the work we've done is making the process as simple and fool-proof as possible. The method was finished prior to even starting on the app.

Try it and let me know how it goes.

 

[Keith_W]

Thanks for posting this thread. Also needs @mitchco and @UliBru to weigh in.

I will type out my interpretation of what @joentell said in this video. The difficulty is that the video is conversational, and I am trying to preserve the meaning in written form. So I can't type out exactly what he said.

- Prior to discussion of his new method of making a target curve, there is a lengthy preamble about target curves, and how it is inappropriate to impose a known "target curve" like the Harman curve on a system. He points out that two different speakers can be equalized to measure the same at the MLP, but they won't sound the same because the major influence is the directivity of the speaker. So far, this seems to be the same point that Dr. Toole made. If you equalize two speakers to measure the same nearfield, they will come a lot closer to sounding the same at the MLP, even if they do not measure the same.

@Floyd Toole has already weighed in in post #4 of this thread and pointed out that "a steady-state room curve - which is what is measured at the listening position - is not a “target”, it is a “result". This is dominated by the off-axis radiation, so a nearfield MMM will not capture it. I have not seen a response from @joentell yet about this point either in this thread, or in that video.

At 11:40 in the video, there is a description of how a customized target curve for your room is made:

- The difference between a measurement of the speaker's anechoic response (here approximated by a nearfield MMM, or maybe an outside measurement), and the MLP measurement of the speaker is the room's transfer function.
- If you make the target the room transfer function, i.e. the delta between the nearfield and the MLP measurement, it will flatten the nearfield response of the speaker.
- You could take a nearfield measurement of the speaker and invert it, and this would flatten the speaker. But this won't correct the bass at the MLP.
- Therefore the best way to correct the bass and flatten the response of the speaker at the same time is to (1) measure the nearfield response of the speaker, (2) measure the response at MLP, (3) subtract nearfield from MLP, and (4) use this as your target (presumably because it will flatten the response of the speaker above Schroder, and correct the bass at the same time).

I would imagine that an alternative approach could work like this:

- take a nearfield MMM of the speaker (ignoring for the time being, Dr. Toole's point above, and assuming for the time being that a nearfield MMM is "good enough" to use as correction). Use this to create an inversion to flatten the response above Schroder.
- take a MMM at the MLP. Use this to flatten the response of low frequencies below Schroder.
- join the two corrections and load that into your convolver

@joentell for the record, I don't feel that you are plugging your product in this thread, nor do I feel that any of your posts are inappropriate or commercial in nature. I find them refreshing and educational. So please continue! This has the potential to become on of the great threads on ASR.

(Note, I am not arguing against anybody here! I can have no opinion of my own, I do not have the qualifications to debate anybody. I only hope that my understanding of the points made is actually the same as the points being made!)

 

[joentell]

Thanks for posting this thread. Also needs @mitchco and @UliBru to weigh in.

I will type out my interpretation of what @joentell said in this video. The difficulty is that the video is conversational, and I am trying to preserve the meaning in written form. So I can't type out exactly what he said.

- Prior to discussion of his new method of making a target curve, there is a lengthy preamble about target curves, and how it is inappropriate to impose a known "target curve" like the Harman curve on a system. He points out that two different speakers can be equalized to measure the same at the MLP, but they won't sound the same because the major influence is the directivity of the speaker. So far, this seems to be the same point that Dr. Toole made. If you equalize two speakers to measure the same nearfield, they will come a lot closer to sounding the same at the MLP, even if they do not measure the same.

@Floyd Toole has already weighed in in post #4 of this thread and pointed out that "a steady-state room curve - which is what is measured at the listening position - is not a “target”, it is a “result". This is dominated by the off-axis radiation, so a nearfield MMM will not capture it. I have not seen a response from @joentell yet about this point either in this thread, or in that video.

At 11:40 in the video, there is a description of how a customized target curve for your room is made:

- The difference between a measurement of the speaker's anechoic response, and the MLP measurement of the speaker is the room's transfer function.
- If you make the target the room transfer function, i.e. the delta between the nearfield and the MLP measurement, it will flatten the nearfield response of the speaker.
- You could take a nearfield measurement of the speaker and invert it, and this would flatten the speaker. But this won't correct the bass at the MLP.
- Therefore the best way to correct the bass and flatten the response of the speaker at the same time is to (1) measure the nearfield response of the speaker, (2) measure the response at MLP, (3) subtract nearfield from MLP, and (4) use this as your target (presumably because it will flatten the response of the speaker above Schroder, and correct the bass at the same time).

I would imagine that an alternative approach could work like this:

- take a nearfield MMM of the speaker (ignoring for the time being, Dr. Toole's point above, and assuming for the time being that a nearfield MMM is "good enough" to use as correction). Use this to create an inversion to flatten the response above Schroder.
- take a MMM at the MLP. Use this to flatten the response of low frequencies below Schroder.
- join the two corrections and load that into your convolver

@joentell for the record, I don't feel that you are plugging your product in this thread, nor do I feel that any of your posts are inappropriate or commercial in nature. I find them refreshing and educational. So please continue! This has the potential to become on of the great threads on ASR.

(Note, I am not arguing against anybody here! I can have no opinion of my own, I do not have the qualifications to debate anybody. I only hope that my understanding of the points made is actually the same as the points being made!)

Great summary! If only ChatGPT were this good!

I'll address the points in reverse chronological order.

All of this can be done for free in REW. It just takes time. It used to take me 4-6 hours to do a 7.1.4 system, and now less that 30 minutes. I welcome the discussion.

I used to use the method you mentioned, but the issue becomes, where do you splice? Where is the transition frequency, and how do you smoothly transition from one method to the other?

I agree with Dr. Toole's assessment. He was responding to a post that didn't represent the method as well as you just did. The MLP measurement is just the result of how X speaker interacted with Y room at Z listening position. What the target there should be is unknown and dependent on the other variables. I'm a huge fan of his work and my takeaway from his research was that people like a speaker that measures flat anechoic with a smooth/linear off-axis response, even when that speaker is placed in a room and affected by all the typical things. He also has mentioned that we can eventually "hear through" the room. Those are some of the things that got me thinking that the target response is for the speaker, and the room is going to do what it's going to do. There's a bit more nuance to that when it comes to the bass, but I think generally speaking, that's how it goes. Trying to fight the room with EQ is a losing battle.

 

[amirm]

Trying to fight the room with EQ is a losing battle.

But that is what you deliver at the end!

 

[Keith_W]

I used to use the method you mentioned, but the issue becomes, where do you splice? Where is the transition frequency, and how do you smoothly transition from one method to the other?

I personally calculate the Schroder frequency using the formula Fs = 2000 * sqrt (T30/V), and the transition frequency as 4Fs. I know this is an approximation (because I can not calculate my room volume exactly, I have adjoining rooms, open staircases, etc) but the prediction based on this calculation is close enough to what I see in measurements.

I use both REW and Acourate, I haven't thought about how I would smoothly transition to one method to another yet, but i'll try to figure it out I am not worried about taking 6 hours to do the measurement and correction, I am doing this for myself in my own system and I am not offering a professional service where time is money

I agree with Dr. Toole's assessment. He was responding to a post that didn't represent the method as well as you just did. The MLP measurement is just the result of how X speaker interacted with Y room at Z listening position. What the target there should be is unknown and dependent on the other variables. I'm a huge fan of his work and my takeaway from his research was that people like a speaker that measures flat anechoic with a smooth/linear off-axis response, even when that speaker is placed in a room and affected by all the typical things. He also has mentioned that we can eventually "hear through" the room. Those are some of the things that got me thinking that the target response is for the speaker, and the room is going to do what it's going to do. There's a bit more nuance to that when it comes to the bass, but I think generally speaking, that's how it goes. Trying to fight the room with EQ is a losing battle.

What concerns me is Dr. Toole's point that a nearfield MMM does not actually capture the off-axis radiation of the speaker. So it is not ideal. Of course, 99% of ASR does not have access to an anechoic chamber of a Klippel so we must make do with what we have. And only some of us have the ability to take the speakers outside to measure (sadly I am not one of them). The vast majority are forced to measure speakers in situ. What I would really like to know is how close a correction based on a nearfield MMM comes to ideal.

 

[joentell]

I personally calculate the Schroder frequency using the formula Fs = 2000 * sqrt (T30/V), and the transition frequency as 4Fs. I know this is an approximation (because I can not calculate my room volume exactly, I have adjoining rooms, open staircases, etc) but the prediction based on this calculation is close enough to what I see in measurements.

I use both REW and Acourate, I haven't thought about how I would smoothly transition to one method to another yet, but i'll try to figure it out I am not worried about taking 6 hours to do the measurement and correction, I am doing this for myself in my own system and I am not offering a professional service where time is money



What concerns me is Dr. Toole's point that a nearfield MMM does not actually capture the off-axis radiation of the speaker. So it is not ideal. Of course, 99% of ASR does not have access to an anechoic chamber of a Klippel so we must make do with what we have. And only some of us have the ability to take the speakers outside to measure (sadly I am not one of them). The vast majority are forced to measure speakers in situ. What I would really like to know is how close a correction based on a nearfield MMM comes to ideal.

I'm not sure I understand the comment about not being able to capture the off-axis response using MMM. If I place a mic 30° horizontally off-axis to the point of reference, that response captures the response you might expect being 30° off-axis. I can spin on a motorized platter while doing a MMM moving the mic vertically up and down and when looking at the difference between that and a LW (10° vertical and 30° horizontal MMM) it sure looks like the LW Directivity Index.

I think it's important to note that this method is for determining the target curve where we do apply smoothing and extreme precision isn't the goal for this app. For speaker design, there's no replacing the accuracy of a Klippel NFS. The thing is, it takes 10 seconds to get about 60 samples using the MMM. The advantage of MMM is repeatability. The downside is there is no phase information. That's why we don't claim to do any impulse response correction. Dirac can do the time domain stuff. We are just determining the target frequency response.

 

[joentell]

But that is what you deliver at the end!

I'm not sure what you mean exactly in response to me saying, "Trying to fight the room with EQ is a losing battle."

I like to think we're working with the room, not against it. For example, if there's extra boundary gain that can be used to extend the bass, that's useful and we can just cut what isn't necessary.

An example of fighting the room would be trying to boost a null. It makes more sense to save the energy that would go into boosting a cancellation and use the headroom for something else.

There are methods of fighting the room, ie. Dirac ART, but that's not the purpose of this app.

 

[Keith_W]

I'm not sure I understand the comment about not being able to capture the off-axis response using MMM. If I place a mic 30° horizontally off-axis to the point of reference, that response captures the response you might expect being 30° off-axis. I can spin on a motorized platter while doing a MMM moving the mic vertically up and down and when looking at the difference between that and a LW (10° vertical and 30° horizontal MMM) it sure looks like the LW Directivity Index.

I see. In other words, you need to know the dispersion of your speakers and then make sure you move your microphone sufficiently to cover the radiation pattern when you are doing the MMM. That makes sense.

However, in the first video I didn't see you waving the mic to cover vertical dispersion. Is this not needed?

I think it's important to note that this method is for determining the target curve where we do apply smoothing and extreme precision isn't the goal for this app. For speaker design, there's no replacing the accuracy of a Klippel NFS. The thing is, it takes 10 seconds to get about 60 samples using the MMM. The advantage of MMM is repeatability. The downside is there is no phase information. That's why we don't claim to do any impulse response correction. Dirac can do the time domain stuff. We are just determining the target frequency response.

Got it I was planning to use Acourate to do everything anyway, except that I would replace the target curve in Acourate with a new target curve that I will derive from your suggestion.

(Edit) this thread also needs @OCA

 

[PHD]

Thank you @amirm
I just want to respect this place by not promoting my app here. I understand that by answering questions, it can be seen as such, but it is not my intention.

I actually think that ASR is a great place to have a dedicated owners' thread for both the upcoming Magic Beans app and the Spatial audio calibration toolkit. This way members could comment and suggest new features which could contribute to further improvements.

 

[Sokel]

I see. In other words, you need to know the dispersion of your speakers and then make sure you move your microphone sufficiently to cover the radiation pattern when you are doing the MMM. That makes sense.

However, in the first video I didn't see you waving the mic to cover vertical dispersion. Is this not needed?



Got it I was planning to use Acourate to do everything anyway, except that I would replace the target curve in Acourate with a new target curve that I will derive from your suggestion.

An easy way to see how important are the off-axis measurement during the speaker construction is this:

Apart from the dotted lines,all the others have to follow a smooth FR respectively to the direct sound on-axis FR through their radiation BEFORE the speaker is even put in a room.

 

[Deleted member 48726]

I'm not sure what you mean exactly in response to me saying, "Trying to fight the room with EQ is a losing battle."

I like to think we're working with the room, not against it. For example, if there's extra boundary gain that can be used to extend the bass, that's useful and we can just cut what isn't necessary.

An example of fighting the room would be trying to boost a null. It makes more sense to save the energy that would go into boosting a cancellation and use the headroom for something else.

There are methods of fighting the room, ie. Dirac ART, but that's not the purpose of this app.

Exactly. The room mode in bass is a helping hand in SPL and extension. It may boost +20 dB it just needs cutting. And we all know boosting a null is not gonna end well.

 

[PHD]

I just found another interesting video on this subject.

 

[Floyd Toole]

Many people like to use the Harman curve based on the JBL Professional M2 measured in various rooms,

Sorry, this is wrong. Revisit the explanation of the origin of what has been called "the Harman curve" in Figure 12.4 in the 3rd edition. It was based on six highly rated loudspeakers, including the M2. The. five others were cone & dome designs. The Figure also shows predicted vs. measured room curves in a variety of rooms done by different people using different measurement methods. The "Harman" curve is a result, not a target. The advantage of starting with well designed loudspeakers is obvious. Fix the bass and enjoy the music. The increasing availability of spinorama data is a powerful positive influence.

If you have the book, have a look at Section 1.7 discussing adaptation - indicating that what we perceive is not completely predictable from in-situ measurements. We adapt to aspects of the room we listen in - not everything needs "correction". This is the challenge: to determine what needs correction and what should simply be left alone.

Finally, there is the important series of tests described in Section 7.6.2, in which three highly rated, similar sounding, loudspeakers were double-blind evaluated in four very different rooms. The similarities in rating and ranking in the different rooms points up the challenge in the kind and amount of "correction" is needed. There is what seems to be clear evidence of adaptation, including the very different bass responses that were present. These, we now know can be significantly corrected. It would be great if someone could fund more of this kind of research, but it is time consuming, expensive and requires some technical expertise.

In the meantime we have opinions and debates,

 

[JPA]

Please explain in more detail.

If I were to say my ideal target would be whatever the response would be in-room of an ideal speaker that measures flat anechoically with a smooth/linear off-axis response but accounting for the affect of the bass in-room, is that not concrete enough? I don't believe there is a static fixed MLP target that applies to all speakers in all rooms. The target is to make the speaker more ideal taking into account physical limitations. How that speaker interacts with the room is room, speaker, speaker placement, and MLP dependent.

Yes, I agree completely, but that's not what you said in the video. You said you were measuring the target response, which is incorrect. As Floyd says above, the target is where you want to end up, and the measurements are your baseline.

You come across as quite intelligent and articulate in your posts on this forum, but not so much in your video. I think you just need to be more careful and precise in the video when you explain what you're doing.

 

[BlackTalon]

It would be great if someone could fund more of this kind of research, but it is time consuming, expensive and requires some technical expertise.

Out of curiosity, what is the rough estimated cost for the next level of research? Is it a $1 mil, $5 mil or $10 mil undertaking?

 

[Floyd Toole]

Out of curiosity, what is the rough estimated cost for the next level of research? Is it a $1 mil, $5 mil or $10 mil undertaking?

I wouldn't hazard a guess at the moment. The research I discussed was supported by the Canadian government at the National Research Council, in a fully equipped laboratory, including the multiple listening rooms and an anechoic chamber, where all of my fundamental research was done. Years of preparation, preceded those tests, and full-time research scientists conducted the tests. To start a similar investigation from scratch would demand a substantial commitment if the results are to be trusted.

Further to my last post, it is worth pondering the fact that starting with my original JAES research papers in 1985-86, and continuing through the years at the NRCC and later at Harman, double-blind listening tests employing hundreds of loudspeakers and listeners in a variety of rooms, monotonously revealed a preference for loudspeakers with minimal evidence of resonant colorations - requiring flat and smooth on-axis anechoic frequency responses with highest scores awarded to those products with smoothly changing off-axis performance. Obviously, listener judgements were not dominated by the rooms - so long as they had time to adapt to the listening environment - which appears to happen quite quickly. Even speech intelligibility improves after a few minutes in a room. Adaptation is real, and starting with well-designed loudspeakers a distinct advantage.

 

[Absolute]

Sorry, this is wrong. Revisit the explanation of the origin of what has been called "the Harman curve" in Figure 12.4 in the 3rd edition. It was based on six highly rated loudspeakers, including the M2. The. five others were cone & dome designs.

Thank you for pointing this out, I was also under the assumption that the curve was based on M2 averages in different rooms.

Btw, I have M2's and have done tons of experiments on all kinds of measurements, MMM near-field, far-field, automatic correction based on both single point and multiple point measurements.
My findings is that single-point measurements is rather useless for both "room-correction" and speaker correction. Here's a demonstration of my speakers with a single point measurement vs MMM compared to the Harman curve;

This is the result of M2's in a highly reflective living room. Hardly likely to surprise you, I'm sure.

With the M2's (they are diy copies since all the parts can be purchased on webshops) I found it impossible to use near-field MMM measurements to find speaker EQ settings and timing/crossover settings because when it looked right at close distance it was off at further distances.
That lead me to believe that MMM is basically a maximum spl average of multiple points in space and therefore can be more or less accurate/useful for speaker correction depending on factors such as distance to speaker relative to driver spacing and distance to reflective surfaces.

I'd love your perspective on my assessment here.

For room EQ in the listening position it works as advertised, here's a comparison between MMM, a 16 point average and single-point measurement;

 

[Blumlein 88]

Seems to me what is missing is someone doing MMM measures in various ways of a speaker or speakers in an anechoic environment and comparing to what a Klippel does. The Klippel is creating the anechoic results from multiple directions without having the anechoic environment.

MMM can within reason be fine for in-room measuring. And we know gated sweeps have a useful connection to anechoic, but they aren't the same thing.

 

[joentell]

Sorry, this is wrong. Revisit the explanation of the origin of what has been called "the Harman curve" in Figure 12.4 in the 3rd edition. It was based on six highly rated loudspeakers, including the M2. The. five others were cone & dome designs. The Figure also shows predicted vs. measured room curves in a variety of rooms done by different people using different measurement methods.

Dr. Toole, I was referring to chapter 11 in the 3rd edition, specifically Figure 11.11 where the JBL M2 was used in 6 different film-sound venues which shows that even the same speaker used in different locations shows a different steady state response. In either case, my full statement was, “Many people like to use the Harman curve based on the JBL Professional M2 measured in various rooms, but they fail to realize that research was descriptive, not prescriptive and wasn't intended for use as a target.” I think you made it pretty clear when you told me over the phone, that it’s not meant to be a target curve. I’ve repeated that statement many times.

Regarding adaptation in section 1.7, I agree and I’ve repeated that many times to others as well. I’ve done experiments myself with binaural recordings I’ve done where the playback is much more convincing when I’m sitting in the same location as where the recording was made vs if I sit in a different room and playback the same recording.

To be clear, when you say in-situ measurement, is that based on a measurement at the listening position or near-field to the speaker?