Equalizing loudspeakers based on anechoic measurements (community project)

[TimVG]

That listening must be done blind.

That's exactly what I'm planning on doing - the PIR optimised filter however sticks out like sore thumb, at least on the 8030C. It's not even an audible improvement over the uncorrected response, no matter what the score says.

It doesn't "suffer" for any of the speakers I've made EQs for. Please have a look at the spreadsheet I posted. Otherwise let's agree to disagree.

But the direct sound does suffer, I've just shown you what the filters do to the response. There is no disagreeing to disagree in this case.

 

[QMuse]

That's exactly what I'm planning on doing - the PIR optimised filter however sticks out like sore thumb, at least on the 8030C. It's not even an audible improvement over the uncorrected response, no matter what the score says.

But the direct sound does suffer, I've just shown you what the filters do to the response. There is no disagreeing to disagree in this case.

It doesn't really matter how the PIR and ON simulated graphs are looking like. All that matters is how you will blindly evaluate uncorrected vs your filters vs flipflop filter vs maiky76 filter compared to the scores of the filters.

It may easilly turn out that you won't be able to tell them apart in a blind test.

 

[TimVG]

It may as well turn out that you won't be able to tell them apart in a blind test.

A treble dip of that width and magnitude is pretty hard to miss.. but we'll see.

 

[Maiky76]

Same result but less extreme than FlipFlops - also lacking energy in the treble. Again after simulation we can see why in terms of direct sound.

View attachment 75374

Awesome!
I have been publishing these EQ in the hope someone would give them a try as I was perplexed by the effect of optimizing the EQ based on the score (i.e. PIR essentially) on the HF and midrange on the ON response. see
https://www.audiosciencereview.com/...urements-community-project.14929/#post-467858

To me it just shows that EQing is not straight forward at all and the directivity of the speaker that can't be EQed is the major concern. The SPDI should be flat on the widest range possible as that is the only way to have both a flat PIR and a flat ON like so:
https://3.bp.blogspot.com/-OB4hm25dXms/XJVUs8TznTI/AAAAAAAAAEA/r6riUCqhZDgJO61yR8uKzWXbHDYsl_CJgCLcBGAs/s1600/Spin+-+Revel+Performa3Be+F228Be.png

if your motivated:

score: 6.7321
297.0, 1.23, 1.04,...
588.0, -0.97, 0.50,...
1673.0, -2.27, 2.52,...
2193.0, 2.10, 0.87,...
4470.0, -0.72, 3.92,...
6962.0, -0.62, 4.47,...
11381.0, 1.30, 2.96,...

You can try it with and without the last PEQ the score is the same really:
6.7366 without the last PEQ a bit better at the HF of the PIR is less boosted.

The difference is that I optimized both for the score AND the flat ON as opposed to just the score.
i.e. make it as flat as possible while not compromising the score too much:

 

[Robbo99999]

Wow, this a really interesting thread, and it's something that I'll be doing on my JBL 308p Mkii once Amir reviews them....he's said he's bought them already, so they're in the pipeline somewhere! I don't fully understand all of the conversations in this thread, but I have a basic understanding of quite a lot of the graphs/points being discussed......my initial strategy would be to EQ my speaker above 500Hz using the on-axis spinorama (I toe them in), I'd EQ that flat, and I may also EQ the speaker in the same way for the area below 500Hz too, but I'd of course need to end up putting more filters on top of that sub 500Hz section to allow for roomEQ so it could be argued that you don't need to EQ the speaker below 500Hz and just EQ the room instead below 500Hz.....at this point I'm undecided. I'll definitely be posting my results in this thread once Amir does get his review done for the 308p Mkii.

 

[TimVG]

Awesome!
I have been publishing these EQ in the hope someone would give them a try as I was perplexed by the effect of optimizing the EQ based on the score (i.e. PIR essentially) on the HF and midrange on the ON response. see
https://www.audiosciencereview.com/...urements-community-project.14929/#post-467858

To me it just shows that EQing is not straight forward at all and the directivity of the speaker that can't be EQed is the major concern. The SPDI should be flat on the widest range possible as that is the only way to have both a flat PIR and a flat ON like so:
https://3.bp.blogspot.com/-OB4hm25dXms/XJVUs8TznTI/AAAAAAAAAEA/r6riUCqhZDgJO61yR8uKzWXbHDYsl_CJgCLcBGAs/s1600/Spin+-+Revel+Performa3Be+F228Be.png

if your motivated:

score: 6.7321
297.0, 1.23, 1.04,...
588.0, -0.97, 0.50,...
1673.0, -2.27, 2.52,...
2193.0, 2.10, 0.87,...
4470.0, -0.72, 3.92,...
6962.0, -0.62, 4.47,...
11381.0, 1.30, 2.96,...

You can try it with and without the last PEQ the score is the same really:
6.7366 without the last PEQ a bit better at the HF of the PIR is less boosted.

The difference is that I optimized both for the score AND the flat ON as opposed to just the score.
i.e. make it as flat as possible while not compromising the score too much:
View attachment 75379

First impression: now this is nice! An improvement over the stock 8030C and a lot better subjectively than your previous filterset. Very close to my own filters in terms of result, now this is worth doing blind.

 

[QMuse]

A treble dip of that width and magnitude is pretty hard to miss.. but we'll see.

Once you saw that dip on the simulated onaxis graph you couldn't possibly miss it when listening, but as soon as that expectation bias is gone it would be much harder to spot it.

Besides, flat on-axis is not the only thing that matters otherwise score would be comprised only of it.

 

[TimVG]

Once you saw that dip on the simulated onaxis graph you couldn't possibly miss it when listening, but as soon as that expectation bias is gone it would be much harder to spot it.

Besides, flat on-axis is not the only thing that matters otherwise score would be comprised only of it.

Well if you read closely I didn't simulate the filters until after I listened to them.. So think of it what you will.

 

[TimVG]

@Maiky76

Would you care to generate filters for the KH80 and M105 based on your adjustment to the 8030C?

 

[edechamps]

Not sure about the "and the higher the slope of the regression"
If the regression perfectly fits the observation then the SM should be 1 regardless of the slope, no?
Especially if the slope becomes infinite (vertical line in the cartesian coordinates) then there is no way to predict the outcome as all a single x value would take all possible outcomes (y values)?

I wouldn't look at it in terms of extremes (such as r²=0, r²=1 or infinite slope), because these involve zero or infinite factors in the r² formula, which makes it impossible to reason about the interaction of the various factors.

Allow me to present a different perspective: r² is defined as 1-(SSres/SStot). If you take a given PIR curve, and you increase its slope (but keep local deviations the same), the SSres of the (newly recomputed) regression will stay the same, but SStot will increase (because PIR points are moving farther away from the mean). The result: r², i.e. SM, increases. Conclusion: local deviations being kept the same, if the overall tilt of the PIR curve increases, SM_PIR will increase.

 

[flipflop]

Very interesting!
Your EQ for the 8030C, 20 EQ points with many tiny sharp adjustments towards NBD: 6.9

My EQ 5 points: score 6.83
PEQ 273.7Hz, 1.17dB, 0.94,...
PEQ 1720.0Hz, -2.27dB, 2.52,...
PEQ 2250.0Hz, 2.25dB, 0.88,...
PEQ 4630.0Hz, -0.97dB, 7.30,...
PEQ 6785.0Hz, -0.73dB, 4.44,...

The NBD is just not the driving factor as I explained, the EQ I design are based on diminishing return based directly on the score.

Our scores are very close considering the fact that you're only using 5 filters. I'm guessing your PIR slope is superior to mine.

But the direct sound does suffer, I've just shown you what the filters do to the response. There is no disagreeing to disagree in this case.

Here's a graph showing the uncorrected on-axis response (brown), the corrected response with my EQ (red), and the target set to the SPL @ 300 Hz (blue):

The corrected response is more accurate from about 400 Hz to 2-4 kHz. The uncorrected response is more accurate for the remaining high frequencies.
The EQ makes a large portion of the response more accurate to a small degree, and diverges from the target to a larger degree for a smaller portion of the response.
We could argue back on forth all day about which one of these graphs is more accurate, but once again I will ask you to look at the results from the spreadsheet:
NBD_ON without EQ: 0.23
NBD_ON with EQ: 0.19

 

[ElNino]

The fourth set, and my favourite, is quite close to the third set in terms of result,but introduces back some of the on-axis energy lost when concentrating solely on the estimated in-room curve (which as we will see in different loudspeakers is quite dependent on overall directivity. Interestingly the estimated in-room response is a happy accident as I did not focus on it when implementing the filters. The idea was to keep the on-axis flat in terms of general trend, and balance the listening window and sidewall reflection (not shown) with this in mind.

Apologies if I've missed a thread somewhere, but I was wondering if you've shared your EQ coefficients anywhere? I'd be interested in trying your fourth set of filters for the M105.

 

[TimVG]

We could argue back on forth all day about which one of these graphs are is more accurate, but once again I will ask you to look at the results from the spreadsheet:

That's all good and well according to your spreadsheet, but which one of us is listening to these speakers at this very moment and is in a better position to judge the audible result?

 

[TimVG]

Apologies if I've missed a thread somewhere, but I was wondering if you've shared your EQ coefficients anywhere? I'd be interested in trying your fourth set of filters for the M105.

Here your are. High Shelf filter @17500 is optional.

 

[TimVG]

First blind results (with thanks to my lovely partner for her help)

8030C uncorrected against flipflop filterset and my own optimised filters.

I was able to identify which was which quite swiftly and accurately, depending on what was playing. Anything involving high treble (e.g. hi-hats) quickly gave away flipflop's filterset - there's just a lack of air and treble there - sorry!

On the uncorrected 8030C the lack of presence on trumpets compared to the other filters was obvious once I knew what to look for.

I switched around Flipflop's filter for Maiky's 2nd filter version, and here things get interesting, as I was not able to distinguish it with total certainty compared to my own filters, let alone provide a definite preference. After my workout session I'm planning on giving them a 2nd try.

TBC

 

[QMuse]

The difference is that I optimized both for the score AND the flat ON as opposed to just the score.
i.e. make it as flat as possible while not compromising the score too much:

IMO that is the best strategy in pretty much every case.

 

[TimVG]

I gave up trying to compare Maiky's 2nd filterset to my own, I could not discern them with absolute certainty after almost 45 min of testing.

I finally loaded both filtersets into REW to compare them one to one.. Here's the listening window comparison

In general above the transition frequency there is never more than 0.5dB difference which is extremely close. Please note no filters were added <300Hz during the listening test.

 

[Robbo99999]

I gave up trying to compare Maiky's 2nd filterset to my own, I could not discern them with absolute certainty after almost 45 min of testing.

I finally loaded both filtersets into REW to compare them one to one.. Here's the listening window comparison

View attachment 75397

In general above the transition frequency there is never more than 0.5dB difference which is extremely close. Please note no filters were added <300Hz during the listening test.

Yeah, I'm pretty sure no one could distinguish those 2 EQ curves, they're practically identical! I would say it doesn't matter which of those you use. If I'd have to pick one I'd probably pick the blue one though, a little more bass and to me looks slightly flatter in the treble.

 

[aarons915]

My general thoughts on EQ so far are that we probably can't use the preference algorithm to make a "perfect" Spinorama. I think the best you can do is make a few good filters like Tim has done that prioritizes the LW then the ER or PIR curve, then a compromise between them and let your listening tests determine which filter sounds best. I think Tim is onto something regarding the directivity of the speakers and maybe trying to force a certain curve in the ER/PIR curve is doing more harm than good in some cases. I think it's much simpler to just go by the general advice of Dr. Toole which is to make the Listening window neutral and the off-axis as smooth as possible. Actually if we simply make the LW neutral and let the speakers directivity dictate the off-axis response, I bet that would be best in many cases. The exception being something like the R3 where a peak develops off-axis and it is actually a problem in listening tests. Some things look bad on a graph but don't really do much when EQ'ing them, someone alluded to that earlier with the filter set of 20-30 filters, which is way too many in my opinion. I prefer to smooth the curves to 1/6 and apply low Q broadband adjustments and I usually don't need more than 5 filters to fix any issues, I would also leave the last octave alone entirely.

I think @TimVG is in a unique position to do some more testing on the ideal directivity though with the Genelec and Neumann speakers. What I think would be interesting is comparing the best EQ of each speaker to each other blind and seeing if there is any preference between a gently rising directivity and constant directivity speaker.

 

[pierre]

Spoiler: Genelec 8030C

Filter 1: ON PK Fc 355 Hz Gain 0.3 dB Q 10.000
Filter 2: ON PK Fc 540 Hz Gain 0.7 dB Q 25.000
Filter 3: ON PK Fc 585 Hz Gain -0.4 dB Q 12.000
Filter 4: ON PK Fc 693 Hz Gain -1.3 dB Q 1.800
Filter 5: ON PK Fc 782 Hz Gain -0.5 dB Q 25.000
Filter 6: ON PK Fc 1105 Hz Gain 0.5 dB Q 25.000
Filter 7: ON PK Fc 1665 Hz Gain 0.6 dB Q 20.000
Filter 8: ON PK Fc 1701 Hz Gain -1.8 dB Q 1.962
Filter 9: ON PK Fc 1745 Hz Gain -0.6 dB Q 25.000
Filter 10: ON PK Fc 2700 Hz Gain -0.5 dB Q 20.000
Filter 11: ON PK Fc 2793 Hz Gain 4.7 dB Q 1.133
Filter 12: ON PK Fc 3239 Hz Gain -3.4 dB Q 1.000
Filter 13: ON PK Fc 3380 Hz Gain 0.3 dB Q 10.000
Filter 14: ON PK Fc 5570 Hz Gain 0.4 dB Q 10.000
Filter 15: ON PK Fc 6680 Hz Gain -0.3 dB Q 8.000
Filter 16: ON PK Fc 8132 Hz Gain -1.6 dB Q 1.000
Filter 17: ON PK Fc 9500 Hz Gain 0.5 dB Q 4.000
Filter 18: ON PK Fc 11250 Hz Gain 0.5 dB Q 7.000
Filter 19: ON PK Fc 14870 Hz Gain -1.0 dB Q 1.000
Filter 20: ON PK Fc 18200 Hz Gain 0.9 dB Q 3.000

Spoiler: Neumann KH80

Filter 1: ON PK Fc 381 Hz Gain -0.7 dB Q 12.000
Filter 2: ON PK Fc 588 Hz Gain 0.8 dB Q 11.000
Filter 3: ON PK Fc 638 Hz Gain -1.0 dB Q 10.000
Filter 4: ON PK Fc 709 Hz Gain -0.3 dB Q 12.000
Filter 5: ON PK Fc 778 Hz Gain -0.7 dB Q 18.000
Filter 6: ON PK Fc 1183 Hz Gain 0.6 dB Q 14.000
Filter 7: ON PK Fc 1540 Hz Gain 0.6 dB Q 8.000
Filter 8: ON PK Fc 1780 Hz Gain 1.4 dB Q 9.000
Filter 9: ON PK Fc 2060 Hz Gain 1.5 dB Q 8.000
Filter 10: ON PK Fc 2330 Hz Gain 1.3 dB Q 15.000
Filter 11: ON PK Fc 2635 Hz Gain 0.9 dB Q 25.000
Filter 12: ON PK Fc 2930 Hz Gain 0.8 dB Q 20.000
Filter 13: ON PK Fc 3232 Hz Gain 0.5 dB Q 24.000
Filter 14: ON PK Fc 3460 Hz Gain -0.4 dB Q 10.000
Filter 15: ON PK Fc 5600 Hz Gain 0.8 dB Q 10.000
Filter 16: ON PK Fc 6500 Hz Gain -0.5 dB Q 10.000
Filter 17: ON PK Fc 9750 Hz Gain -1.0 dB Q 4.500
Filter 18: ON PK Fc 10940 Hz Gain 0.4 dB Q 10.000
Filter 19: ON PK Fc 12650 Hz Gain -0.7 dB Q 5.000
Filter 20: ON PK Fc 16000 Hz Gain -2.6 dB Q 1.400

Spoiler: Revel M105

Filter 1: ON PK Fc 382 Hz Gain -0.4 dB Q 18.000
Filter 2: ON PK Fc 445 Hz Gain 0.8 dB Q 5.000
Filter 3: ON PK Fc 505 Hz Gain -1.0 dB Q 5.000
Filter 4: ON PK Fc 604 Hz Gain 1.1 dB Q 7.000
Filter 5: ON PK Fc 660 Hz Gain -0.8 dB Q 10.000
Filter 6: ON PK Fc 917 Hz Gain -0.4 dB Q 3.000
Filter 7: ON PK Fc 1540 Hz Gain -0.4 dB Q 15.000
Filter 8: ON PK Fc 1650 Hz Gain 0.5 dB Q 10.000
Filter 9: ON PK Fc 1760 Hz Gain -0.5 dB Q 15.000
Filter 10: ON PK Fc 2000 Hz Gain -0.4 dB Q 15.000
Filter 11: ON PK Fc 2361 Hz Gain -0.9 dB Q 1.000
Filter 12: ON PK Fc 2451 Hz Gain -0.4 dB Q 20.000
Filter 13: ON PK Fc 2555 Hz Gain 0.9 dB Q 22.000
Filter 14: ON PK Fc 3300 Hz Gain -0.3 dB Q 5.000
Filter 15: ON PK Fc 4894 Hz Gain -2.1 dB Q 1.482
Filter 16: ON PK Fc 4950 Hz Gain 0.6 dB Q 18.000
Filter 17: ON PK Fc 5650 Hz Gain -0.4 dB Q 6.000
Filter 18: ON PK Fc 8050 Hz Gain 0.5 dB Q 4.000
Filter 19: ON PK Fc 10515 Hz Gain -0.8 dB Q 1.000
Filter 20: ON PK Fc 13800 Hz Gain -0.4 dB Q 1.700

pierre has yet to calculate the post-EQ scorse for these speakers, so I can't tell you what they are yet. Of the 4 speakers Maiky76 posted EQs for, it's actually only the PreSonus E5 XT I've got a post-EQ score of. Sorry about that.
I'm still confident those filters yield a substantial improvement, though.


All my EQs are from 300 Hz and up.

all fliflop@‘s eq are on github Here ...