MMM approach and a new calibration app (magic beans)

[joentell]

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.

I know. I've done it more than a few times.

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



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

Correct

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?

Correcting the MLP to target the room response does both at the same time as long.

 

[sigbergaudio]

Correcting the MLP to target the room response does both at the same time as long.

If the room response is defined by the difference between MMM MLP and MMM NF, I'd say maybe.

If you only did a single MLP measurement, you'd definitely get things corrected above Schroeder that you shouldn't change. But given you do MMM I guess most of this is hopefully evened out.

Would be interesting to see some before/after graphs (single measurement MLP of individual speaker channels) of the system in action in a few different rooms, including information about both the room and the speakers.

 

[joentell]

If the room response is defined by the difference between MMM MLP and MMM NF, I'd say maybe.

That's what we're using to determine it.

If you only did a single MLP measurement, you'd definitely get things corrected above Schroeder that you shouldn't change. But given you do MMM I guess most of this is hopefully evened out.

Would be interesting to see some before/after graphs (single measurement MLP of individual speaker channels) of the system in action in a few different rooms, including information about both the room and the speakers.

If you have a calibrated mic and REW, you can try it. I made a post showing my RTA settings that I use.

 

[Blumlein 88]

Here is the thread I couldn't remember. Using two microphones to remove bass nodes. Perhaps this could be used for the lower frequencies as several measurements have shown issues. Then splice it to whatever Magic Beans is doing above Schroeder range.

Using two microphones to remove room modes from bass measurements

I found this article and wonder if anyone has experience with this technique of making quasi anechoic measurements in regular rooms down to 10Hz. https://audioxpress.com/article/measuring-loudspeaker-spl-response-and-harmonic-distortion-at-low-frequencies/ The microphone array costs 700€ and...

www.audiosciencereview.com

 

[Sokel]

As I did a lot of junk in previous posts,this time I took my time also reading the paper @Blumlein 88 posted in the other thread.
Here's the results.no EQ,no nothing:


80cm (red) vs MLP (green)


Same with psy and overlayed at 1/48.


...and a lows only (as better as I could as everything was playing) just for the fun of it.

(rename the .zip file to .mdat after you download it)

 

[Keith_W]

Hi all. After spending a few days investigating this method, I have come to a rather sad conclusion: I don't think it would work for this speaker, and in this room.

As I have explained in my previous post, the objective of this method is to correct the speakers so that they are flat with a nearfield measurement, which will then produce a falling Harman-like curve at MLP. The correction is obtained by subtracting the NF response from MLP.

I posted these curves yesterday. My "nearfield" measurement was taken 30cm from the speaker. Once it is subtracted from the MLP, the result is the Delta curve which shows a rising frequency response.

If you use this as your target curve, it should correct frequencies above Schroder to be flat, which it does. This is a really pleasing result, and it shows that @joentell's method works as intended - it really does flatten the nearfield response.

(As an aside, I am having issues with tweeter volume at the moment (see separate discussion), which is why there is a downwards step in the right tweeter (green line) above the XO frequency of 5kHz.)

However, when the verification measurement was taken from the MLP, I had a rising response in the upper frequencies. It sounds as bad as you would expect, see previous post for listening impressions.

This prompted me to ask myself why my speaker should produce a rising frequency response at MLP instead of a falling one, which one would expect. I exchanged some PM's with @joentell, and he suggested that horn speakers like mine have an optimum distance where the sound "comes together", sort of a focal point if you like.

So today, I repeated the measurements. Just FYI this is the measurement setup:

You can see the size of the speaker. I lay a tape measure on the floor and proceeded to take MMM's at various distances. The volume of the speaker was NOT adjusted between all these runs of measurements. The result is below:

The idea behind doing this is to find the ideal distance to take "nearfield" measurements from. What I am looking for is the rate of change between curves to stop changing. If two subsequent measurements at different distances show no difference (apart from volume), then the "nearfield" is somewhere between those two measurements.

This is what I see:
- the volume of the tweeter (at about 15kHz) remains the same even as the microphone gets further away from the speaker.
- the volume of the rest of the spectrum (100Hz - about 5kHz) falls as the mic gets further away from the speaker. This is expected behaviour.

The combination of constant tweeter volume + falling midrange horn + woofer volume means that if the speaker is equalized to be flat at "nearfield", it will produce a rising frequency response as seen in the third graph. My reasoning is this: if it is flat at "nearfield", as you get further away, the woofer + midrange horn drop off at a faster rate than the tweeter. This produces the rising response.

Conclusion: this horn loaded tweeter does not appear to behave the same way as a conventional tweeter in that it does not seem to lose much energy as the mic gets further away from it. I suspect that in this room, this speaker might even "come together" at a point behind the MLP, in other words there needs to be more distance between the MLP and the speaker.

So sadly, much as I love @joentell's idea, I don't think that this system is a good candidate for it.

 

[sigbergaudio]

Hi all. After spending a few days investigating this method, I have come to a rather sad conclusion: I don't think it would work for this speaker, and in this room.

As I have explained in my previous post, the objective of this method is to correct the speakers so that they are flat with a nearfield measurement, which will then produce a falling Harman-like curve at MLP. The correction is obtained by subtracting the NF response from MLP.

I posted these curves yesterday. My "nearfield" measurement was taken 30cm from the speaker. Once it is subtracted from the MLP, the result is the Delta curve which shows a rising frequency response.

If you use this as your target curve, it should correct frequencies above Schroder to be flat, which it does. This is a really pleasing result, and it shows that @joentell's method works as intended - it really does flatten the nearfield response.

(As an aside, I am having issues with tweeter volume at the moment (see separate discussion), which is why there is a downwards step in the right tweeter (green line) above the XO frequency of 5kHz.)

However, when the verification measurement was taken from the MLP, I had a rising response in the upper frequencies. It sounds as bad as you would expect, see previous post for listening impressions.

This prompted me to ask myself why my speaker should produce a rising frequency response at MLP instead of a falling one, which one would expect. I exchanged some PM's with @joentell, and he suggested that horn speakers like mine have an optimum distance where the sound "comes together", sort of a focal point if you like.

So today, I repeated the measurements. Just FYI this is the measurement setup:

View attachment 339025

You can see the size of the speaker. I lay a tape measure on the floor and proceeded to take MMM's at various distances. The volume of the speaker was NOT adjusted between all these runs of measurements. The result is below:

View attachment 339027
The idea behind doing this is to find the ideal distance to take "nearfield" measurements from. What I am looking for is the rate of change between curves to stop changing. If two subsequent measurements at different distances show no difference (apart from volume), then the "nearfield" is somewhere between those two measurements.

This is what I see:
- the volume of the tweeter (at about 15kHz) remains the same even as the microphone gets further away from the speaker.
- the volume of the rest of the spectrum (100Hz - about 5kHz) falls as the mic gets further away from the speaker. This is expected behaviour.

The combination of constant tweeter volume + falling midrange horn + woofer volume means that if the speaker is equalized to be flat at "nearfield", it will produce a rising frequency response as seen in the third graph. My reasoning is this: if it is flat at "nearfield", as you get further away, the woofer + midrange horn drop off at a faster rate than the tweeter. This produces the rising response.

Conclusion: this horn loaded tweeter does not appear to behave the same way as a conventional tweeter in that it does not seem to lose much energy as the mic gets further away from it. I suspect that in this room, this speaker might even "come together" at a point behind the MLP, in other words there needs to be more distance between the MLP and the speaker.

So sadly, much as I love @joentell's idea, I don't think that this system is a good candidate for it.

Out of curiosity, if you measure your speakers (left and right individually) at ear height at MLP (so not MMM but just a single measurement) without correction, what does that look like?

 

[OCA]

I can make a video about it eventually. Here's a measurement I just found that I took using MMM overlaid over a Klippel NFS one. This is prior to me realizing that UMIK-1 is slightly out of calibration (treble is too hot) from being dropped one too many times. I should be getting a Cross Spectrum Labs UMIK-1 soon.

The term 'Kali' implies a connection to Linux hacking!

 

[joentell]

Hi all. After spending a few days investigating this method, I have come to a rather sad conclusion: I don't think it would work for this speaker, and in this room.

As I have explained in my previous post, the objective of this method is to correct the speakers so that they are flat with a nearfield measurement, which will then produce a falling Harman-like curve at MLP. The correction is obtained by subtracting the NF response from MLP.

I posted these curves yesterday. My "nearfield" measurement was taken 30cm from the speaker. Once it is subtracted from the MLP, the result is the Delta curve which shows a rising frequency response.

If you use this as your target curve, it should correct frequencies above Schroder to be flat, which it does. This is a really pleasing result, and it shows that @joentell's method works as intended - it really does flatten the nearfield response.

(As an aside, I am having issues with tweeter volume at the moment (see separate discussion), which is why there is a downwards step in the right tweeter (green line) above the XO frequency of 5kHz.)

However, when the verification measurement was taken from the MLP, I had a rising response in the upper frequencies. It sounds as bad as you would expect, see previous post for listening impressions.

This prompted me to ask myself why my speaker should produce a rising frequency response at MLP instead of a falling one, which one would expect. I exchanged some PM's with @joentell, and he suggested that horn speakers like mine have an optimum distance where the sound "comes together", sort of a focal point if you like.

So today, I repeated the measurements. Just FYI this is the measurement setup:

View attachment 339025

You can see the size of the speaker. I lay a tape measure on the floor and proceeded to take MMM's at various distances. The volume of the speaker was NOT adjusted between all these runs of measurements. The result is below:

View attachment 339027
The idea behind doing this is to find the ideal distance to take "nearfield" measurements from. What I am looking for is the rate of change between curves to stop changing. If two subsequent measurements at different distances show no difference (apart from volume), then the "nearfield" is somewhere between those two measurements.

This is what I see:
- the volume of the tweeter (at about 15kHz) remains the same even as the microphone gets further away from the speaker.
- the volume of the rest of the spectrum (100Hz - about 5kHz) falls as the mic gets further away from the speaker. This is expected behaviour.

The combination of constant tweeter volume + falling midrange horn + woofer volume means that if the speaker is equalized to be flat at "nearfield", it will produce a rising frequency response as seen in the third graph. My reasoning is this: if it is flat at "nearfield", as you get further away, the woofer + midrange horn drop off at a faster rate than the tweeter. This produces the rising response.

Conclusion: this horn loaded tweeter does not appear to behave the same way as a conventional tweeter in that it does not seem to lose much energy as the mic gets further away from it. I suspect that in this room, this speaker might even "come together" at a point behind the MLP, in other words there needs to be more distance between the MLP and the speaker.

So sadly, much as I love @joentell's idea, I don't think that this system is a good candidate for it.

I’ve really enjoyed our discussion via PM where we’ve discussed critical distance and directivity of your speakers. And you’re correct in saying that these might not be a proper candidate for this method, but also keep in mind that the app does have things in place to prevent over-correction that doing it manually does not provide. The idea was to test the concept of whether MLP > Room response = flat NF, and I think this shows that. But, NF for your speakers based on the size of the waveguide, is much further back than a typical speaker. This still helps me in that, I think 1X the distance from the lowest contributing driver to the top of the waveguide is probably a safe guideline for a NF MMM.

What’s implemented into our app is a limit on boost on the high frequencies to prevent overcorrection. That is not shown by using the method directly without any limits.

 

[Keith_W]

Out of curiosity, if you measure your speakers (left and right individually) at ear height at MLP (so not MMM but just a single measurement) without correction, what does that look like?

Here you go, before and after. No prizes for guessing which one is "before" and which one is "after".

 

[sigbergaudio]

Here you go, before and after. No prizes for guessing which one is "before" and which one is "after".

View attachment 339046

Thank you. The "after" graphs here is way to aggressively corrected for my taste. This is what I'm a bit curious about with this magic beans system. Would be interesting with a system that corrected closer to what one would do manually (which is not much at all above Schroeder).

 

[joentell]

Here you go, before and after. No prizes for guessing which one is "before" and which one is "after".

View attachment 339046

I find your speaker fascinating. I would really like to get an idea of the directivity index on that. The idea behind not over correcting with DSP is that we don’t correct what is not correctable, for example, if there’s a directivity mismatch.

It seems to me that what’s considered NF for the large horn specifically is different from what would be considered NF for the rest of the speaker. Therefore, it might be interesting to EQ them separately. This is only because we’re doing this manually. With most auto-EQ, including MB, it will limit the amount of boost, especially at higher frequencies. The public release of the MB app would not apply that large boost above 5kHz.

 

[Keith_W]

I’ve really enjoyed our discussion via PM where we’ve discussed critical distance and directivity of your speakers. And you’re correct in saying that these might not be a proper candidate for this method, but also keep in mind that the app does have things in place to prevent over-correction that doing it manually does not provide. The idea was to test the concept of whether MLP > Room response = flat NF, and I think this shows that. But, NF for your speakers based on the size of the waveguide, is much further back than a typical speaker. This still helps me in that, I think 1X the distance from the lowest contributing driver to the top of the waveguide is probably a safe guideline for a NF MMM.

Just FYI:

The curve at 100cm would look somewhere in between the 90cm and 150cm measurement. One of my friends (speaker designer) mentioned to me that when designers model their designs, they can optimize for listening position, and he thinks that the designers of my speaker may have optimized for a further distance. The MLP is only 3.5m away from the speakers, and I did not bother taking MMM's further than 200cm from the speaker because by then "nearfield" would be almost at the listening position, and I did not think it made sense.

I did write to Acapella a few years back asking for their Spinorama but they never replied my email. Who knows, they may be annoyed at me for modifying their speaker beyond all recognition (including swapping out their woofer, bypassing the passive crossovers, etc). There is a lot that I do not know about my speaker and this exercise has been totally worth it, even if it did not result in a new correction curve that I could use.

Thank you. The "after" graphs here is way to aggressively corrected for my taste. This is what I'm a bit curious about with this magic beans system. Would be interesting with a system that corrected closer to what one would do manually (which is not much at all above Schroeder).

Yep. I mentioned that upthread, I even included a quote from Toole who said that if I have a perfect looking curve, it is likely that something inappropriate has been done. Well, I have a perfect looking curve. Ergo, I must have done something inappropriate. Which was the motivation for me to investigate this method.

BTW I think it is a brilliant idea! Kudos to @joentell for coming up with it.

 

[joentell]

Just FYI:

View attachment 339049
View attachment 339050

The curve at 100cm would look somewhere in between the 90cm and 150cm measurement. One of my friends (speaker designer) mentioned to me that when designers model their designs, they can optimize for listening position, and he thinks that the designers of my speaker may have optimized for a further distance. The MLP is only 3.5m away from the speakers, and I did not bother taking MMM's further than 200cm from the speaker because by then "nearfield" would be almost at the listening position, and I did not think it made sense.

I did write to Acapella a few years back asking for their Spinorama but they never replied my email. Who knows, they may be annoyed at me for modifying their speaker beyond all recognition (including swapping out their woofer, bypassing the passive crossovers, etc). There is a lot that I do not know about my speaker and this exercise has been totally worth it, even if it did not result in a new correction curve that I could use.



Yep. I mentioned that upthread, I even included a quote from Toole who said that if I have a perfect looking curve, it is likely that something inappropriate has been done. Well, I have a perfect looking curve. Ergo, I must have done something inappropriate. Which was the motivation for me to investigate this method.

BTW I think it is a brilliant idea! Kudos to @joentell for coming up with it.

I don’t care what anyone says, that’s a cool looking speaker! I especially think that plasma tweeter looks crazy cool! But maybe there’s something to the idea that there are speakers that are too large for the room. Maybe the large horns do require listening from a further distance. I appreciate you helping me with this, because these edge cases help me improve the method for others. Now, a Klipsch RF7II horn has nothing on these! If we can not over-correct your speaker, then we won’t have a problem with the Klipsch!

 

[ernestcarl]

Conclusion: this horn loaded tweeter does not appear to behave the same way as a conventional tweeter in that it does not seem to lose much energy as the mic gets further away from it. I suspect that in this room, this speaker might even "come together" at a point behind the MLP, in other words there needs to be more distance between the MLP and the speaker.

There are two horns… I don’t remember anymore but I am guessing the smaller gold one is producing the rising HF?

Increasing variation below 700Hz looks very normal as well as the increasing swings below 200Hz.

It’s only how the HF retains similar SPL past 3m that’s unusual.

You could simply just look at the effect of the EQ at all distances/positions anyhow as that would demonstrate how balanced you can make the system everywhere. 30 cm is definitely too close for this multiway to get an overall NF picture — though it probably would be fine for some coaxes.

 

[twelti]

Thoughts:

Low freq: I think you should just use the room response at MLP for low frequencies. At low frequencies, other than the fact that the source/room/receiver system is highly dependent on source/receiver locations, the speaker and room tend to act as a single unified system (lumped element). So no difference between direct, early and late reflections, as you get at higher frequencies (due to temporal masking etc.). Just a frequency response that is heard. Even if you use MM technique, you are still essentially measuring in some arbitrary location in the room where the speaker is and same for the arbitrary measurement for the listener. That means you get an arbitrary correction when you take the difference. If you get close enough to the speaker to start to get "direct" sound at low frequencies, I wonder if it would be very sensitive to exactly where you measured. If you measured truly close, as in a near field measurement around 1/8" from the woofer, maybe you get a good estimate of sound power output, but still, I just think at low frequencies you can measure at MLP, and correct that directly. One caveat, if there are multiple seating locations, not just one MLP, then measuring near field at the woofer might make sense. A resonance in the source will be in the response for all seating positions, so correcting it separately might make sense. I would still go for a single measurement very close to the woofer cone (towards the edge of the dust cap if memory serves), instead of a MM average.

Mid/High freq: Maybe I misunderstood the method. If you take a standard "good" speaker with flat on-axis and typical DI curve, and measure near field you get a flat curve, or maybe closer to listening window curve which should still be pretty flat for a reference speaker. Then the result of the calibration would be a flat in-room curve at the MLP which for typical speakers will have too much high frequency. I do agree that measuring close to the speaker might be interesting (Harman had an AVP processor back a few years that did that using the remote controller). I suppose that doing some weighted correction which attempts to flatten direct field response and reverberent field response in some more intelligent manner might be interesting.

If you are proposing some truly novel, a listening test of some sort (double blind etc.) is a good idea!

 

[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.

Hi Joe. OK as requested. NF MMMs for Left Centre and Right speakers:-

I might have another go with the right speaker. It is the one with the most close reflective surfaces so could just be that.
I found REW's A/B 'regularisation' feature useful for ignoring any nulls in the original measurement.
Filters were all calculated in DL live - I imported the MB target into the DL app.

You asked about listening impressions. Actually I found the difference surprisingly subtle. Perhaps the MB approach was a fraction clearer? I didn't level match either so take with a pinch of salt. I have always found turning the volume up a notch is the best sound quality tweak!

 

[Sokel]

There are two horns… I don’t remember anymore but I am guessing the smaller gold one is producing the rising HF?

Increasing variation below 700Hz looks very normal as well as the increasing swings below 200Hz.

It’s only how the HF retains similar SPL past 3m that’s unusual.

You could simply just look at the effect of the EQ at all distances/positions anyhow as that would demonstrate how balanced you can make the system everywhere. 30 cm is definitely too close for this multiway to get an overall NF picture — though it probably would be fine for some coaxes.

Next time I'll bring measurements from big horns (2 m).
It would be nice to know what and where to measure,so suggestions would be appreciated.

 

[joentell]

Thoughts:

Low freq: I think you should just use the room response at MLP for low frequencies. At low frequencies, other than the fact that the source/room/receiver system is highly dependent on source/receiver locations, the speaker and room tend to act as a single unified system (lumped element). So no difference between direct, early and late reflections, as you get at higher frequencies (due to temporal masking etc.). Just a frequency response that is heard. Even if you use MM technique, you are still essentially measuring in some arbitrary location in the room where the speaker is and same for the arbitrary measurement for the listener. That means you get an arbitrary correction when you take the difference. If you get close enough to the speaker to start to get "direct" sound at low frequencies, I wonder if it would be very sensitive to exactly where you measured. If you measured truly close, as in a near field measurement around 1/8" from the woofer, maybe you get a good estimate of sound power output, but still, I just think at low frequencies you can measure at MLP, and correct that directly. One caveat, if there are multiple seating locations, not just one MLP, then measuring near field at the woofer might make sense. A resonance in the source will be in the response for all seating positions, so correcting it separately might make sense. I would still go for a single measurement very close to the woofer cone (towards the edge of the dust cap if memory serves), instead of a MM average.

Mid/High freq: Maybe I misunderstood the method. If you take a standard "good" speaker with flat on-axis and typical DI curve, and measure near field you get a flat curve, or maybe closer to listening window curve which should still be pretty flat for a reference speaker. Then the result of the calibration would be a flat in-room curve at the MLP which for typical speakers will have too much high frequency. I do agree that measuring close to the speaker might be interesting (Harman had an AVP processor back a few years that did that using the remote controller). I suppose that doing some weighted correction which attempts to flatten direct field response and reverberent field response in some more intelligent manner might be interesting.

If you are proposing some truly novel, a listening test of some sort (double blind etc.) is a good idea!

What an honor to have you chime in!

For the subs, we use an average of the room response from the other speakers as a target response for the subs. The idea is to augment the sound while retaining the natural tonality of the speakers played in that room (room effects included.)

For the mid/highs, we do aim to get something like a listening window response. For that ideal speaker, the measurement at the MLP and the "room transfer function" or room response would be nearly identical which means that the difference between the two would lead to a flat line, meaning no need for correction to that ideal speaker. I agree 100% that a flat response at MLP is not what we want.

 

[joentell]

Hi Joe. OK as requested. NF MMMs for Left Centre and Right speakers:-
View attachment 339108View attachment 339109 View attachment 339110

I might have another go with the right speaker. It is the one with the most close reflective surfaces so could just be that.
I found REW's A/B 'regularisation' feature useful for ignoring any nulls in the original measurement.
Filters were all calculated in DL live - I imported the MB target into the DL app.

You asked about listening impressions. Actually I found the difference surprisingly subtle. Perhaps the MB approach was a fraction clearer? I didn't level match either so take with a pinch of salt. I have always found turning the volume up a notch is the best sound quality tweak!

Thank you for taking the time to do this.