EQ targets for wide dispersion speakers

[ernestcarl]

not sure I understand your graphs here.

Basically, it just shows that MMM, LR vector avg (multiple sweeps), and RMS avg (same multiple sweeps) all yield similar results 2m at the MLP. Also, the measurement at 2m (~15-20 degrees off-axis) resembles the measurement I took at 15cm (15+30 degrees directivity curves avg'd).

 

[dasdoing]

I thought you meant anechoic measurement taken in the nearfield (orange trace). So what you mean is at 2m distance at my MLP using FDW 5 like this?

View attachment 115992

yes. that was the whole idea with exporting the MP version. correcting those dips which are caused by delay of those frequencies.

it is quite diferent from the quasi-anechoic meassurement then

 

[ernestcarl]

yes. that was the whole idea with exporting the MP version. correcting those dips which are caused by delay of those frequencies.

it is quite diferent from the quasi-anechoic meassurement then

True. This is the reason why I do not want to over-correct using measurements taken at 2m. Only small adjustments to mainly match the differences between left and right channels -- and mainly cuts. If I filled every single dip and peak found in the RMS and/or Vector averaged curves aggressively (2m), the MMM (2m) and nearfield anechoic directivity curve (15cm) measurements are also all drastically altered.

 

[richard12511]

If there's EQ involved above a few hundred hertz, it should be speaker correction based on anechoic measurements. If not possible to better the speaker that way, buy better speakers.

I honestly wonder how much these conclusions depend on the limitations of the DRC systems of the time. Would full range correction from a system like Audiolense show a different subjective result? Subjectively, I do slightly prefer my 8351b with full range correction, though the difference is very small.

 

[Absolute]

I honestly wonder how much these conclusions depend on the limitations of the DRC systems of the time. Would full range correction from a system like Audiolense show a different subjective result? Subjectively, I do slightly prefer my 8351b with full range correction, though the difference is very small.

These results were before the Audiolense/Accurate's of today, so it would be nice to have those factors included.
But saying that, there's no black magic involved in those programs. Run a single-point measurement correction with any neutral speaker and do a before vs after MMM measurement from around the listening position.

Is it better or worse after correction? Better? Then all is well. Worse? Oops.

For me I can't get better sound with full-range correction than I get with only low-frequency correction in my room. Here's before and after MMM;

Red is after Audiolense vs before. Needless to say the tonality is way off after correction.
Sounds warm and cuddly, but far from realistic. I don't think there's a clear answer here, in some cases it might be better and in some it might be worse. So many possibilities to screw it up with these modern aids that you really need to know what you're doing.

After a couple of hundred hours with Dirac and Audiolense I gave it up.

 

[thewas]

After a couple of hundred hours with Dirac and Audiolense I gave it up.

I also used to use Acourate quite intensely in the past 5 years but nowadays just correct if needed based on anechoic measurements making LW kind of flat and only few room mode peaks below 200 Hz.

 

[richard12511]

After a couple of hundred hours with Dirac and Audiolense I gave it up.

Your target curve looks much different than the actual response of your speaker, and looks like it's changing the actual tonality of the speaker?

I honestly have a hard time hearing the difference between limited and full range comparison, and I'm not totally convinced that I even can. Tonality doesn't change at all(at least with my default curve) to my ears, and it just tightens up a bit and image clarity improves ever so slightly with full range comparison. I'll see if I can get some measurements, but from memory, mine show less difference between full range and limited range. Looking at your two measurements, I can see how the full range correction might sound worse(since it's changing the tonality of the neutral speaker).

 

[Absolute]

The target response was actually rather close to the measured response, just tried to deal with the SBIR and modal peaks and valleys.

It's a difficult scenario for room correction when lp is close to walls.

 

[test1223]

I recently assembled a pair of Linkwitz Pluto - inspired 3-way speakers. Like the original Plutos, they are omnidirectional up to about 3khz. Out of the box, the frequency response was pretty funky, so I've been using REW to design a parametric EQ for them.

With a typical (Harman-like target) roll off of about 2db/octave starting from 2khz, they sound pretty dark in comparison. I compared them side by side with typical sealed box speakers- definitely subjectively much darker. With a flat or even slightly up-tilted response from 2khz onwards, they start to sound more natural.

Has anyone else experienced this? Do the dispersion characteristics of a speaker affect the ideal frequency response target?

Hey,

almost all literature is assuming none omnidirectional speaker as an object to equalize. The Pluto speakers are omnidirectional. If you use an omnidirectional speaker at a higher listening distance the sound is dominated by the later reflections which provide in a good scenario a diffuse sound field.

The tonal perception of direct sound is different than the perception of diffuse sound due to the head related transfer functions.

Therefore the room target curve of an omnidirectional speaker with higher listening distance should be different than with a very short listening distance.

There isn't much literature of this issue jet. But the finding from here might help.

Best
Thomas

 

[mitchco]

So many possibilities to screw it up with these modern aids that you really need to know what you're doing.

This right here

 

[ernestcarl]

Red is after Audiolense vs before. Needless to say the tonality is way off after correction.

I cannot help but wonder as to what happened there... There is so much uneven wide boosts applied throughout the upper bass, low-mids and mids. Esp. those last two — if the speakers were already flat in the anechoic the corrected response shouldn’t have been so drastically altered after correction. And I mean the change here looks really, really drastic.

 

[Absolute]

I cannot help but wonder as to what happened there... There is so much uneven wide boosts applied throughout the upper bass, low-mids and mids. Esp. those last two — if the speakers were already flat in the anechoic the corrected response shouldn’t have been so drastically altered after correction. And I mean the change here looks really, really drastic.

Certainly, and this is what happens (or can happen) when you base any correction on a single point measurement. Dirac is even worse when you do that, it looks like this;

Kii Three vs Kii Three with Dirac Single point measurement correction

I didn't know about MMM method back then unfortunately, but we can of course just imagine the same broad shelf filter in the low mids. Sounded horrible.

More interesting example is my testing of a porous bass trap (just one package of insulation wrapped in plastic placed in the corner next to the righ speaker vs EQ in the low frequency range.

First without basstrap and no EQ;

With basstrap and no EQ;

Frequency response;

Ok, so one package of insulation just placed unopened in one corner does this much. But since EQ is effective, let's see what we can achieve with EQ instead. No basstrap vs only EQ;

Wow. The time-domain is vastly improved even if no bass trap is present. Below 30 hz I don't really notice the difference in the time-domain, so I didn't bother to EQ that part away as that's "free capacity".

One question remains, what if I place the basstrap back in the corner and then modify the EQ to suit? Will it be even better?
First we try the above EQ and place back the trap before we adjust the EQ to have something to compare it too;

Unmodified EQ (suited to non trap) + basstrap in corner

Modified EQ to account for basstrap in corner;

Wow again, it gets worse in the time domain! Seems like we need to decide whether or not to go for maximum performance in the frequency domain OR in the time domain.

The point of all this with regards to EQ targets?
EQ and room interaction is complicated and should really be explored from different viewpoints. Apply a peak filter here and you will introduce a resonance to the speaker, room or both. The time domain is so important that Dr. Toole stated We should avoid resonances at all costs!
In this case here we see that a simple trap raises the response in the SBIR range of 100-200 hz by around 3 dB with the added bonus of taking away issues in the time domain. Doing this with EQ only will introduce more room ringing as well as introducing a resonance to the speaker.
Below 100 hz that won't matter as we're not sensitive enough to pick that up, but once we go above 100 hz or so we quickly become very sensitive to the time-domain.

https://asa.scitation.org/doi/abs/10.1121/1.4908217?journalCode=jas&

Here's what happens in the time domain if you apply a peak filter;

Before vs after peak filter at 5 kHz

My conclusion is that we're asking quite a lot from automatic room correction systems when we just randomly pick a spot to put the microphone and let the system correct the whole room in both time and frequency domain based on that little information.
It stands to reason that it will be very variable how successful that turns out.

These new programs are so powerful that you can easily turn crisp crust into cheesecake before you blink. At the very least get someone like @mitchco to take a look at the measurement files and set up the software to fix what's reasonable and leave what isn't.

 

[akarma]

When EQ i avoid any peak filter at all. Only notch filter

MMM measurement from sweetspot is the best solution for EQ in ordinary room.

 

[richard12511]

Certainly, and this is what happens (or can happen) when you base any correction on a single point measurement. Dirac is even worse when you do that, it looks like this;

Kii Three vs Kii Three with Dirac Single point measurement correction
View attachment 116770

I didn't know about MMM method back then unfortunately, but we can of course just imagine the same broad shelf filter in the low mids. Sounded horrible.

More interesting example is my testing of a porous bass trap (just one package of insulation wrapped in plastic placed in the corner next to the righ speaker vs EQ in the low frequency range.

First without basstrap and no EQ;
View attachment 116772

With basstrap and no EQ;
View attachment 116773

Frequency response;

View attachment 116774

Ok, so one package of insulation just placed unopened in one corner does this much. But since EQ is effective, let's see what we can achieve with EQ instead. No basstrap vs only EQ;

View attachment 116776

Wow. The time-domain is vastly improved even if no bass trap is present. Below 30 hz I don't really notice the difference in the time-domain, so I didn't bother to EQ that part away as that's "free capacity".

One question remains, what if I place the basstrap back in the corner and then modify the EQ to suit? Will it be even better?
First we try the above EQ and place back the trap before we adjust the EQ to have something to compare it too;

Unmodified EQ (suited to non trap) + basstrap in corner
View attachment 116778

Modified EQ to account for basstrap in corner;
View attachment 116779

Wow again, it gets worse in the time domain! Seems like we need to decide whether or not to go for maximum performance in the frequency domain OR in the time domain.

The point of all this with regards to EQ targets?
EQ and room interaction is complicated and should really be explored from different viewpoints. Apply a peak filter here and you will introduce a resonance to the speaker, room or both. The time domain is so important that Dr. Toole stated We should avoid resonances at all costs!
In this case here we see that a simple trap raises the response in the SBIR range of 100-200 hz by around 3 dB with the added bonus of taking away issues in the time domain. Doing this with EQ only will introduce more room ringing as well as introducing a resonance to the speaker.
Below 100 hz that won't matter as we're not sensitive enough to pick that up, but once we go above 100 hz or so we quickly become very sensitive to the time-domain.

https://asa.scitation.org/doi/abs/10.1121/1.4908217?journalCode=jas&

Here's what happens in the time domain if you apply a peak filter;

Before vs after peak filter at 5 kHz
View attachment 116786
View attachment 116787

My conclusion is that we're asking quite a lot from automatic room correction systems when we just randomly pick a spot to put the microphone and let the system correct the whole room in both time and frequency domain based on that little information.
It stands to reason that it will be very variable how successful that turns out.

These new programs are so powerful that you can easily turn crisp crust into cheesecake before you blink. At the very least get someone like @mitchco to take a look at the measurement files and set up the software to fix what's reasonable and leave what isn't.

Interesting, I've never actually tried EQing off a single point measurement. The corrections I'm seeing are much less dramatic, so I'm guessing that's why? For someone like me who mostly cares about a single point, is there a reason to use single point EQ?

I'm interested to try it now, though. I think the least I've ever used is 10 measurements spread over the width of my wingspan, depth of one arm, and height a little taller than my upper body.

 

[Absolute]

Interesting, I've never actually tried EQing off a single point measurement. The corrections I'm seeing are much less dramatic, so I'm guessing that's why? For someone like me who mostly cares about a single point, is there a reason to use single point EQ?

I'm interested to try it now, though. I think the least I've ever used is 10 measurements spread over the width of my wingspan, depth of one arm, and height a little taller than my upper body.

Indeed, and personally I find that even Audiolense sounds far better when correcting for many measurement points even if it's not really meant for it.

In my opinion it's silly to correct for a single point in space because no single point in space carries with it enough information to accurately predict spatial quality. In my case there's a huge dip around 70 hz right were my head would be, but 20 cm up or down it turns into a peak. We listen to low frequencies with our whole body, so what should I correct for?
And going further, if I correct for a dip at the LP that is a huge peak 1 meter in front, did I make the sound better or worse in the room?
Can we hear a huge room resonance in the room even if it's not present right were you sit? In my experience, yes you sometimes can.
Putting 10 dB into a local dip will increase the peak somewhere else with 10 dB.

I believe that we should at least measure several points within 0,5 meter up, down, front and back (if possible) and 1 meter to each side from where you sit to avoid stupid stuff. Most of us will likely have at least one wall within close proximity to the listening area and that will create huge variety in local response. Example where I'm measuring exactly where my head is when I rest it against the back of the sofa vs 1 meter out.
That's a 20+ dB difference at 350 hz with no EQ. What happens if I correct for that head-resting point?

This is not an attack on room correction software, it's just sharing my experience and understanding of how complicated acoustics can be in a given situation. My findings are simple, room correction sounds far better when correcting for many points in space and sounds best when limited to below 500 hz - if a decent speaker is used.

 

[ernestcarl]

I believe that we should at least measure several points within 0,5 meter up, down, front and back (if possible) and 1 meter to each side from where you sit to avoid stupid stuff.

Which is easy to do with MMM. For the time domain information throughout the entire frequency range there is still no substitute for taking multiple sweeps at the listening position(s). Ideally REW should have a feature that would allow one to apply filters directly to multiple curves and/or IRs at the same time so you can easily visualize changes simultaneously within one screen -- they could add it as a "Pro Upgrade" feature. Of course, there are some proprietary/in-house-built software from loudspeaker manufacturers that can do this already.

 

[mitchco]

Re: single versus multiple analysis measurements - know how your DRC software is intended to be used: https://www.audiosciencereview.com/...-altitude-jbl-sdp-75.13095/page-9#post-621360

 

[ernestcarl]

Re: single versus multiple analysis measurements - know how your DRC software is intended to be used: https://www.audiosciencereview.com/...-altitude-jbl-sdp-75.13095/page-9#post-621360

Haven't found much to really disagree with your more advanced DRC approach... However, I'm also quite satisfied with the results I've been able to obtain using multi-point measurements and MMM within REW and in combination with rePhase. I probably won't be going back to only using single-point analysis and EQ myself. Either way the learning curve is steep -- and not to mention still ongoing.

 

[Absolute]

Which is easy to do with MMM. For the time domain information throughout the entire frequency range there is still no substitute for taking multiple sweeps at the listening position(s). Ideally REW should have a feature that would allow one to apply filters directly to multiple curves and/or IRs at the same time so you can easily visualize changes simultaneously within one screen -- they could add it as a "Pro Upgrade" feature. Of course, there are some proprietary/in-house-built software from loudspeaker manufacturers that can do this already.

You can average many measurements and then auto-eq that in REW. I compared this yesterday with MMM and it really turns out very similar. I don't have the measurements on my phone right now, but I believe the difference between the two after EQ (measured by MMM) was within +- 1-2 dB or so below 300 hz.

MMM just saves so much time for the same result that I see no reason to go through the pain of measuring 30+ points to average them.

It could still be beneficial to take single-point measurements below 100 hz to check waterfall plots and make adjustments to EQ based on that, but I know it's debated whether or not it makes too much of a difference once the frequency response is good. For me I find that it makes a difference how we achieve the summed response, but YMMV.

 

[ernestcarl]

You can average many measurements and then auto-eq that in REW. I compared this yesterday with MMM and it really turns out very similar. I don't have the measurements on my phone right now, but I believe the difference between the two after EQ (measured by MMM) was within +- 1-2 dB or so below 300 hz.

MMM just saves so much time for the same result that I see no reason to go through the pain of measuring 30+ points to average them.

It could still be beneficial to take single-point measurements below 100 hz to check waterfall plots and make adjustments to EQ based on that, but I know it's debated whether or not it makes too much of a difference once the frequency response is good. For me I find that it makes a difference how we achieve the summed response, but YMMV.

I do have some measurement comparisons here:

The differences are not big. Because the HF does change in level -- depending on which method is used -- one probably needs a separate "target curve". The bass also will differ in level if it's a summed or single channel capture.

Due to the "quickness" of MMM, it's certainly the first thing I always use now to quickly EQ a speaker. I still do a combination auto-generated filters and manual adjustments whichever the method -- esp. if there's more than one listening position. But the reason I still use full sweeps at all is because it provides much deeper information for room acoustics, plus speaker time and distortion analysis which you cannot acquire any other way.