MMM approach and a new calibration app (magic beans)

[PH73]

Interesting. I would be interested in seeing the NF MMM response with the Dirac curve on, and another one with MB. I am also curious what your subjective listening experience is like using both the MB target response curve and the default one Dirac provides

Dirac On, NF mmm is in the final chart.

Measured MB corrected NF mmm will have to wait a couple of days until I get a chance to load the target into DL and send the filters into the AVR. That will require a bit of file manipulation I think. In theory it should be flat.
Same goes for listening impressions ( in mono). Your app would certainly speed things up if I wanted to do multiple speakers.
Btw I am not using the default Dirac target curve. It is a classic smooth ski- slope type curve (as you can see) but created manually to be fairly close to the mlp measurement. I intended it to be more about smoothing the response than forcing a whole different shape on the room. Perhaps that is why my DL filters look quite close to the implied MB ones.

 

[joentell]

Dirac On, NF mmm is in the final chart.

Measured MB corrected NF mmm will have to wait a couple of days until I get a chance to load the target into DL and send the filters into the AVR. That will require a bit of file manipulation I think. In theory it should be flat.
Same goes for listening impressions ( in mono). Your app would certainly speed things up if I wanted to do multiple speakers.
Btw I am not using the default Dirac target curve. It is a classic smooth ski- slope type curve (as you can see) but created manually to be fairly close to the mlp measurement. I intended it to be more about smoothing the response than forcing a whole different shape on the room. Perhaps that is why my DL filters look quite close to the implied MB ones.

I like that idea of smoothing the natural response as long as you have some good speakers. If the NF response has that that HF roll-off NF, does that still correct to you? I guess it depends on your speaker design and how reflective your room is.

 

[PH73]

I need to find time to explore this more. My current thoughts:-

My speakers are DIY 4 way active crossover so pretty much flat on-axis, gated.

NF mmm in situ is not flat.

In the bass NF picks up room modes that are common to MLP, so the MB approach does a good job filtering them out. The overall level of bass targeted by MB is sensitive to the distance at which the NF mmm is taken. This could be particularly true for my speakers where I think driver integration becomes a problem if I go too close.

In the mid range, MB works well. It doesn't try to fill in the holes (like DL does) at 140 and 375 that are only present in the farfield. These could well be room effects that we 'listen through'. Something to test and explore.

In the high frequency, the NF mmm rolls off whereas the on axis stays flat. Typical increasing directivity. This is likely quite sensitive to how the mmm is taken - how wide and tall the movement range is. MB tries to correct for this by boosting the top octave. I am unsure how desirable this is. It probably depends on how toed in the speakers are. My hearing doesn't go much above 14 khz anyway.

When I get time I will try again for at least both main l&r. I think I will use the bigger distance but tighter movement range on the NF mmm. I might play around with different smoothing too. Then I will finally do some listening !

 

[joentell]

I need to find time to explore this more. My current thoughts:-

My speakers are DIY 4 way active crossover so pretty much flat on-axis, gated.

NF mmm in situ is not flat.

In the bass NF picks up room modes that are common to MLP, so the MB approach does a good job filtering them out. The overall level of bass targeted by MB is sensitive to the distance at which the NF mmm is taken. This could be particularly true for my speakers where I think driver integration becomes a problem if I go too close.

In the mid range, MB works well. It doesn't try to fill in the holes (like DL does) at 140 and 375 that are only present in the farfield. These could well be room effects that we 'listen through'. Something to test and explore.

In the high frequency, the NF mmm rolls off whereas the on axis stays flat. Typical increasing directivity. This is likely quite sensitive to how the mmm is taken - how wide and tall the movement range is. MB tries to correct for this by boosting the top octave. I am unsure how desirable this is. It probably depends on how toed in the speakers are. My hearing doesn't go much above 14 khz anyway.

When I get time I will try again for at least both main l&r. I think I will use the bigger distance but tighter movement range on the NF mmm. I might play around with different smoothing too. Then I will finally do some listening !

I appreciate you taking the time to test! Thank you!

Happy 2024 everyone!

 

[Deleted member 48726]

Just for laughs, here's a NF MMM measurement of a Monolith THX Compact Satellite speaker placed on the dash of my van (in orange), and here's the response with the engine on and with road noise while driving (in purple). No wonder a system can sound good while parked with the engine off, and while driving, the bass sounds so weak. It needs to overcome the road noise! I saved a preset with that ridiculous bass rise as my target and it actually sounds normal while driving.View attachment 337928
Maybe that's why this Mercedes SUV I measured had this response with a nearly 20dB bass rise? I thought it was so odd when I saw that.
View attachment 337930

That's why you want really beastly subwoofers and amplifiers in car audio to have proper (audible) bass.

 

[sigbergaudio]

@joentell

Soooooo, there are a number of somewhat confusing explanations in this thread, and there's also somewhat longish videos trying to explain it (that I didn't watch all the way).

But the way I understand this the difference between this and a typical room correction system is that:

A typical room correction system has either a flat or sloped default target, while your system has the nearfield response of the speakers as target. Right?

I would be interested to hear how aggressive the correction is.

So how much does it correct (+/-dB)?
How precise (smoothing)?
In which frequency range does it operate (below schroeder, in the schroeder transition, full range)?

 

[Keith_W]

A typical room correction system has either a flat or sloped default target, while your system has the nearfield response of the speakers as target. Right?

It has the "room transfer function" as the target. This is derived from subtracting the nearfield MMM from the MLP MMM.

 

[Blumlein 88]

It has the "room transfer function" as the target. This is derived from subtracting the nearfield MMM from the MLP MMM.

That is just a round about way of saying the same thing.

 

[joentell]

That is just a round about way of saying the same thing.

This isn’t the same as what @sigbergaudio mentioned about using the NF measurement as the target response. I’ve seen some room corrections that measure the MLP position and attempt to make it measure like the NF measurement (as a target) to attempt to remove the room’s influence. I’ve tried a few of them and just sounded bad as you might expect.

This is very different from using the room’s response as the target for MLP.

 

[Blumlein 88]

This isn’t the same as what @sigbergaudio mentioned about using the NF measurement as the target response. I’ve seen some room corrections that measure the MLP position and attempt to make it measure like the NF measurement (as a target) to attempt to remove the room’s influence. I’ve tried a few of them and just sounded bad as you might expect.

This is very different from using the room’s response as the target for MLP.

So what is the rooms transfer function if it isn't the NF response as modified by the room at the MLP? I'm misunderstanding something.

 

[sigbergaudio]

This isn’t the same as what @sigbergaudio mentioned about using the NF measurement as the target response. I’ve seen some room corrections that measure the MLP position and attempt to make it measure like the NF measurement (as a target) to attempt to remove the room’s influence. I’ve tried a few of them and just sounded bad as you might expect.

This is very different from using the room’s response as the target for MLP.

It has the "room transfer function" as the target. This is derived from subtracting the nearfield MMM from the MLP MMM.

So the MMM nearfield response is the target?

EDIT: Or are you somehow figuring out how that speaker "should" sound in the listening position if the room didn't interfer?

 

[thewas]

I have been using a nearfield (at around 30-50 cm from loudspeaker baffle) listening window MMM approach (around +-30° hor and +-15° ver) for the correction of the region above 500 Hz for several years and it gives usually good results as it matches also the average corresponding sweeps at that region well:

Below 500 Hz I correct separately for left and right loudspeaker based on MMM at the listening position to a flat down to 200 Hz and after 3-.5 dB rising target but I find also the here presented approach of measuring the room response for the bass target interesting.

 

[Keith_W]

So the MMM nearfield response is the target?

EDIT: Or are you somehow figuring out how that speaker "should" sound in the listening position if the room didn't interfer?

See this post.

As I understand it:

Harman "target" (more like ideal Harman "result") at the MLP = Flat Speaker + Room Transfer Function.

If we want a Harman result, we need a Flat Speaker and we need to know the Room Transfer Function.

For "Flat Speaker", @joentell proposes that we obtain this by doing a nearfield MMM, which will capture direct and off-axis radiation. If we want to, we can simply do an inversion of the response to correct the speaker to flat.

To derive "Room Transfer Function", his other proposal is to subtract the nearfield MMM from the MLP MMM.

If we make the target curve the "Room Transfer Function", this should correct the speaker to flat and correct any frequencies below Schroder simultaneously.

I have been constantly experimenting every day since I read the first post. I have avoided posting more results because I am having a lot of trouble, and frankly my results are unbelievable. By this I mean that the result I am seeing is the opposite of what I am expecting to see. For example, if we take a flat speaker, put it in the room, then measure at the MLP, we expect to see a falling frequency response, right? Well, this is what I have:

Rising frequency response at the MLP! I have no idea why my speakers are behaving this way. Maybe the fact that the tweeters are horn loaded might have something to do with it. I have looked at the ETC, waterfall, etc. to try to explain why I am seeing such a treble boost at the MLP and I have come up with nothing.

 

[sigbergaudio]

All these explanations have so many words, I started watching these youtube videoes and there are analogies and bla bla bla, isn't it possible to say exactly what it is doing?

@Keith_W So based on your explanation there are two things this is doing:

1) Attempt to correct the response of the actual speaker and make it as flat as possible based on nearfield measurements, covering the listening window.
2) Attempt to correct of the in-room response below Schroeder based on measurements at the main listening position.

Correct? Or no?


One thing is what the software is trying to achieve, and the second is how. If you guys separated those, I suspect it would be easier to explain.


If we look at a completely manual process, one would typically get good results with the following process:

1) Purchase as good a set of speakers one could afford
2) Correct for the room below schroeder, somewhat agressively below 100hz, and more carefully above 100hz.
3) Optionally adjust tonality to your preference

Most room correction software fail on the emulation of step 2, because they correct way too agressively and way too high in frequency. Is this new software different?

 

[ernestcarl]

See this post.

As I understand it:

Harman "target" (more like ideal Harman "result") at the MLP = Flat Speaker + Room Transfer Function.

If we want a Harman result, we need a Flat Speaker and we need to know the Room Transfer Function.

For "Flat Speaker", @joentell proposes that we obtain this by doing a nearfield MMM, which will capture direct and off-axis radiation. If we want to, we can simply do an inversion of the response to correct the speaker to flat.

To derive "Room Transfer Function", his other proposal is to subtract the nearfield MMM from the MLP MMM.

If we make the target curve the "Room Transfer Function", this should correct the speaker to flat and correct any frequencies below Schroder simultaneously.

I have been constantly experimenting every day since I read the first post. I have avoided posting more results because I am having a lot of trouble, and frankly my results are unbelievable. By this I mean that the result I am seeing is the opposite of what I am expecting to see. For example, if we take a flat speaker, put it in the room, then measure at the MLP, we expect to see a falling frequency response, right? Well, this is what I have:

View attachment 338750

Rising frequency response at the MLP! I have no idea why my speakers are behaving this way. Maybe the fact that the tweeters are horn loaded might have something to do with it. I have looked at the ETC, waterfall, etc. to try to explain why I am seeing such a treble boost at the MLP and I have come up with nothing.

Apply aggressive frequency dependent windowing to your measurement at the MLP. The rise in energy may be from the room reflections.

 

[ernestcarl]

2) Correct for the room below schroeder, somewhat agressively below 100hz, and more carefully above 100hz.

Even with MMM, one needs to be cautious because some of the “corrections” or compromises the speaker designer made could well be for perfectly logical reasons.

The “prophile” coaxes esp. the smallest P8 from Fulcrum Acoustic would not be easy to correctly balance without having access to the full set of on- and off-axis curves.

Some of those decisions are not only based on the technical side of measurements but also of aesthetic ones as well during listening.

If there is a directivity error like due to horn diffraction, say, the RX series between 5-9 kHz the MMM nearfield will show it as wide dip that should not be flattened out. One will need swept sine measurements to confirm.

 

[Keith_W]

Well, I have some results. I think I am moving forward, apart from some unexpected issues with my system and room.

Every time I make a working directory with Acourate, I write myself a note so that I can understand my motivation for doing this project and the steps that I took. This is a copy of my text file. Skip reading this if you wish, it's only for Acourate users:

The purpose of this iteration is to correct the main speakers and subwoofers separately.

For the main speakers:
1. MMM taken NF of L/R with the subs turned off.
2. MMM taken at MLP of L/R with the subs turned off.
3. NF subtracted from MLP (Delta MLP-NF NoSubs L/R) to give room transfer function with FD-Functions Magnitude Difference

For the subs:
1. MMM taken NF of L/R subs with mains turned off.
2. MMM taken MLP of L/R subs with mains turned off.
3. NF subtracted from MLP (Delta MLP-NF SubOnly L/R) to give room transfer function with FD-Functions Magnitude Difference

To create target curve:
1. Open Delta NoSubs L/R and Delta SubOnly L/R and level match the measurements with TD-Functions Gain
2. FD-Functions - Magnitude Split n Join with split freq of 80Hz
3. FDW 15/15
4. Phase Extraction 20Hz - 21kHz, save as Target L/R

Acourate has a limitation that MMM recordings can not go through Macro 4 (Filter Generation). Therefore we will process MMM and SPS separately.
1. Create MMM subfolder, record fullrange MMM, pink noise 1kHz, 1/12 smoothing. Rename MMLeft/Right to Pulse48L/Pulse48R.
1.1 Run Macro 1 (Amplitude preparation) normally
1.2 Skip Macro 2 (Target Curve Design). Instead, copy Target L/R from previous step into your working directory.
1.3 Run Macro 3 (Inversion) with the following setting: Mono, UNCHECK "Use default Target", and specify Pulse48L/R and TargetL/R. You have to run Macro 3 twice, once for left, again for right.
1.3 Copy Pulse48Linv / Pulse48Rinv to SPS subfolder.

2. Create SPS subfolder, record fullrange Pulse48L/R.
2.1 Do not run any macros.
2.2 Proceed to Macro 4 once Pulse48Linv / Pulse48Rinv copied from MMM subfolder
---> FINISH

Once the filters were created, I went on to do the verification measurements.

This is a nearfield MMM of left and right speakers. Ignore the horrible bass for now, will comment on it later. It is nice to see that it corrects pleasingly flat, so the "room transfer function" correction seems to work as intended.

There right speaker (green curve) has a step at 5kHz. I am working on sorting out the issue, which is a bit too involved to discuss in this thread (and will drag it off topic). For now it has something to do with how Acourate reports the result of an SPS vs. MMM. I am chatting with Uli about it.

However, the MLP measurements are quite different:

The top curve (red/green) are MMM's at MLP. Bottom curve (brown/blue) are SPS at MLP. As you can see, the responses are very similar, and quite unusual.

Unexpected result #1: I would have thought that the SPS would look different, because they are windowed to remove reflections, whereas the MMM should capture speaker sound power.

Unexpected result #2: See the rising treble response in both measurements? Somehow, flat nearfield results in rising treble response at the MLP. This is completely opposite of what I would expect and totally confusing. I have never seen this kind of behaviour in the past, because my previous practice involves correcting the response at the MLP to a falling target curve.

Unexpected result #3: The bass has not been corrected with this procedure.

Listening Impressions: As I expected, it sounds horrible. Waaaay too thin sounding. Voices are all top end and sibilant with no body. That "one note" bass is really evident, cellos sound almost comical (all top end with some notes having a huge bass boost).

Conclusions / things to do: This result is unlistenable but it was an interesting experiment, even though I consider it a partial failure. I had some expected results, and some unexpected results. I will need another method to correct below Schroder. I don't think this method works. I need to figure out why my room produces a rising treble response (hopefully others will post their results so I can see the difference between NF and MLP measurements).

 

[joentell]

All these explanations have so many words, I started watching these youtube videoes and there are analogies and bla bla bla, isn't it possible to say exactly what it is doing?

@Keith_W So based on your explanation there are two things this is doing:

1) Attempt to correct the response of the actual speaker and make it as flat as possible based on nearfield measurements, covering the listening window.
2) Attempt to correct of the in-room response below Schroeder based on measurements at the main listening position.

Correct? Or no?


One thing is what the software is trying to achieve, and the second is how. If you guys separated those, I suspect it would be easier to explain.


If we look at a completely manual process, one would typically get good results with the following process:

1) Purchase as good a set of speakers one could afford
2) Correct for the room below schroeder, somewhat agressively below 100hz, and more carefully above 100hz.
3) Optionally adjust tonality to your preference

Most room correction software fail on the emulation of step 2, because they correct way too agressively and way too high in frequency. Is this new software different?

I find it fascinating that many people are finding this so confusing. @Keith_W explained it well in this post and in previous posts.

I think where people are stuck is they seem to think that a target curve is something that is pre-determined prior to taking a single measurement. I propose that it's something that you arrive at after correcting the speaker properly. The method corrects the NF response to a flat and the bass is smoothed based on the MLP response. That's the target. The response curve at the MLP is a result of those corrections.

The other thing people seemed to be confused about is how we use the room response (how the room changed the sound) as a target with which we correct the MLP response in order to achieve the corrections I stated above. I'm simplifying here, but If you had a perfectly flat speaker and a you measured at your MLP, the room will change the measured response. If you make that response the target with which you make a correction of the MLP response, what you'll see is there's no difference between the MLP and the room response, so the corresponding correction to that speaker will be no correction. This is correct because the speaker we started with is already ideal.

The thing that might be tough to grasp and tough to believe as that using the room's transfer function as the target for the MLP response will yield a NF measurement that is flat. It's easier to see with actual measurements and examples from others trying it and showing the results.

 

[joentell]

Well, I have some results. I think I am moving forward, apart from some unexpected issues with my system and room.

Every time I make a working directory with Acourate, I write myself a note so that I can understand my motivation for doing this project and the steps that I took. This is a copy of my text file. Skip reading this if you wish, it's only for Acourate users:



Once the filters were created, I went on to do the verification measurements.

View attachment 338766

This is a nearfield MMM of left and right speakers. Ignore the horrible bass for now, will comment on it later. It is nice to see that it corrects pleasingly flat, so the "room transfer function" correction seems to work as intended.

There right speaker (green curve) has a step at 5kHz. I am working on sorting out the issue, which is a bit too involved to discuss in this thread (and will drag it off topic). For now it has something to do with how Acourate reports the result of an SPS vs. MMM. I am chatting with Uli about it.

However, the MLP measurements are quite different:

View attachment 338768

The top curve (red/green) are MMM's at MLP. Bottom curve (brown/blue) are SPS at MLP. As you can see, the responses are very similar, and quite unusual.

Unexpected result #1: I would have thought that the SPS would look different, because they are windowed to remove reflections, whereas the MMM should capture speaker sound power.

Unexpected result #2: See the rising treble response in both measurements? Somehow, flat nearfield results in rising treble response at the MLP. This is completely opposite of what I would expect and totally confusing. I have never seen this kind of behaviour in the past, because my previous practice involves correcting the response at the MLP to a falling target curve.

Unexpected result #3: The bass has not been corrected with this procedure.

Listening Impressions: As I expected, it sounds horrible. Waaaay too thin sounding. Voices are all top end and sibilant with no body. That "one note" bass is really evident, cellos sound almost comical (all top end with some notes having a huge bass boost).

Conclusions / things to do: This result is unlistenable but it was an interesting experiment, even though I consider it a partial failure. I had some expected results, and some unexpected results. I will need another method to correct below Schroder. I don't think this method works. I need to figure out why my room produces a rising treble response (hopefully others will post their results so I can see the difference between NF and MLP measurements).

If you have a large horn/waveguide, you have to measure from further back. I'm glad you're taking the time to do this because it helps me come up with a procedure to determine the minimum distance required to take a MMM based on the speaker design. A proper distance for NF MMM will differ from speaker to speaker. If the distance between your tweeter and the furthest driver is far, the minimum distance for MMM will need to be further as well. What I think is happening in your case is that you are measuring at a distance that doesn't capture the full response at Nf, but the MLP is due to the distance. Either that or your room is very reflective.

Let me know what you find out. Also, I'm curious to work with you to try this in REW. I'm just very unfamiliar with Accurate. We'll see what we can learn from this. I've seen this before though. It was with a Klipsch RF7II. The method is only as good as the measurements we take. The goal is to figure out how to get the best measurements for all types of speakers.

The way I think about some of these waveguides is I imagine a parabolic dish. There's a certain focal point where the energy is extremely focused. Anywhere but that point will show less energy. Of course, the waveguide's purpose is slightly different, but there will be areas of measurement where the sound is "out of focus." If the speaker is designed so the point of focus where the speakers converge and sum properly are as far back as your MLP, then this method won't work because your MLP and distance required for you NF measurement are nearly identical.

I'm thinking we might have to build some functionality into the app to help people determine the optimal distance for their speaker based on their measured NF response.

 

[sigbergaudio]

I find it fascinating that many people are finding this so confusing. @Keith_W explained it well in this post and in previous posts.

It's normal to try to repeat something to verify it has been understood. Then it is helpful if the one who originally explained confirms that yes, you understood it correctly, or no, that's not it.

The method corrects the NF response to a flat and the bass is smoothed based on the MLP response. That's the target.

This sounds like the same thing I said. If that is the case, it appears I have understood.

The other thing people seemed to be confused about is how we use the room response (how the room changed the sound) as a target with which we correct the MLP response in order to achieve the corrections I stated above. I'm simplifying here, but If you had a perfectly flat speaker and a you measured at your MLP, the room will change the measured response. If you make that response the target with which you make a correction of the MLP response, what you'll see is there's no difference between the MLP and the room response, so the corresponding correction to that speaker will be no correction. This is correct because the speaker we started with is already ideal.

So in other words, a good speaker should not be corrected (above Schroeder).

The thing that might be tough to grasp and tough to believe as that using the room's transfer function as the target for the MLP response will yield a NF measurement that is flat. It's easier to see with actual measurements and examples from others trying it and showing the results.

This is where it becomes a bit confusing. Are you changing the response based on both the NF and the MLP in the same frequency range? Or are you adjusting based on NF above Schroeder and based on MLP below Schroeder?