Preferred Target FR Curve

[AetherDrive]

Hi Everyone!

TLDR; what's your preferred FR Curve and EQ goals when using DSP?

I've built my own pair of speakers of my own design, you can find my write up of it here: Still Haven't Named Them. I stopped messing with them for a bit and am very happy with its sound signature except I found that it had a really bad peak in the 4k range, most music didn't highlight it but when it did it was audible. Long story short, tried Dirac Live to help fix it, which it did, but it completely changed the overall sound, removing some of the dynamic range in the bass and making everything more homogeneous in its presentation. Not an unenjoyable experience but it just feels like something is missing from the mix. I've AB'd tracks with and without the Dirac and have started to manually EQ them in an attempt to keep my preferred sound with some of the benefits from the Dirac's processing. What I've found is that my speakers have room for improvement but I'm not sure where to go beyond flattening out some of the dips and peaks that my FR chart shows. Also, I'm aware that EQ alone won't get me the same effect that Dirac gives, just trying to improve the base and then see where Dirac can take me, if I choose to use it.

I know it's a terribly subjective question and changes based on the speaker but are there any good FR Curve goals other than, make it flat, that you target/find that you prefer? I'm using a Dayton DSP-408 for active xo and eq.

 

[kemmler3D]

are there any good FR Curve goals other than, make it flat, that you target/find that you prefer?

I'd say that this is the most reliable path forward. Once you get it flat, you can start to think about what's missing and what's overdone, and EQ accordingly by hand.

If you want to know about my preferences for deviation from flat, if I go that route: (off the top of my head)


Roughly:
+3 dB from 40-20hz
+2 dB from 40- 60hz
- 2 dB from 80-120hz
- 2 dB between 250 and 500hz
+ 2 dB between 700 and 1500hz
- 2 dB between 2400 and 4000 hz

Flat from there on. So call it a "tight but warm" bass-to-midrange tuning.

But, this is really just how I feel this morning. Get your system to be flat first, and then you can make good decisions on your own "house curve". Keep in mind the off-axis response of your speakers might fight you on certain EQ decisions.

 

[digitalfrost]

At the moment I use this:

0 -40.0
28 -20.0
29 0.26
35 0.26
41 0.26
47 0.25
53 0.25
59 0.25
64 0.24
70 0.23
76 0.22
82 0.21
88 0.2
1000 0.0
20000 -5.0
22000 -40.0
24000 -100.0

Looks like that:

I wrote some software to create these curves, so I can easily tune them in 0.1dB increments. I don't think it makes a lot of sense to compare them, since the effect of the curve will heavily depend on the software used to do the correction and the settings used. Of course also speakers and room.

From what I learned at least for me, there should be some kind of knee point where the curve begins to drop. I've used a linear curve before (well it's a logarithmic scale but you get my point...) and if you use any type of bass boost it will simply put too much energy around 200hz. Example with "correct" bass boost applied +3dB at 10hz and -3dB at 20khz:

So what I'm doing here is stiching multiple things together to make a single curve, because I do not want 200hz to rise just because of the bass boost. Similarly, if there would be no knee to control when the high frequency drop starts, you'd end you with not enough energy around 1khz if you already allow the curve to drop starting say 300hz.

Mockup with a straight line curve (green). This is not good. If you use more bass boost it gets even worse.

In my current curve I have 0.26dB bass boost and you would think one can't hear that, but I can. I guess this comes down not to my golden ears, but to the implementation of my DSP software. There's something happening at 0.26dB boost where it just boosts the bass enough that does not happen at 0.25dB. These things are not perfect and also have to work with limited resolution. Somehow that 0.1dB brings me just over the thresold where the bass becomes nice.

 

[Thomas_A]

Hi Everyone!

TLDR; what's your preferred FR Curve and EQ goals when using DSP?

I've built my own pair of speakers of my own design, you can find my write up of it here: Still Haven't Named Them. I stopped messing with them for a bit and am very happy with its sound signature except I found that it had a really bad peak in the 4k range, most music didn't highlight it but when it did it was audible. Long story short, tried Dirac Live to help fix it, which it did, but it completely changed the overall sound, removing some of the dynamic range in the bass and making everything more homogeneous in its presentation. Not an unenjoyable experience but it just feels like something is missing from the mix. I've AB'd tracks with and without the Dirac and have started to manually EQ them in an attempt to keep my preferred sound with some of the benefits from the Dirac's processing. What I've found is that my speakers have room for improvement but I'm not sure where to go beyond flattening out some of the dips and peaks that my FR chart shows. Also, I'm aware that EQ alone won't get me the same effect that Dirac gives, just trying to improve the base and then see where Dirac can take me, if I choose to use it.

I know it's a terribly subjective question and changes based on the speaker but are there any good FR Curve goals other than, make it flat, that you target/find that you prefer? I'm using a Dayton DSP-408 for active xo and eq.

Can you show the on-axis frequency response with scale 50 dB on the Y-axis? Gated.

 

[ppataki]

I'd say that this is the most reliable path forward. Once you get it flat, you can start to think about what's missing and what's overdone, and EQ accordingly by hand.

If you want to know about my preferences for deviation from flat, if I go that route: (off the top of my head)


Roughly:
+3 dB from 40-20hz
+2 dB from 40- 60hz
- 2 dB from 80-120hz
- 2 dB between 250 and 500hz
+ 2 dB between 700 and 1500hz
- 2 dB between 2400 and 4000 hz

Flat from there on. So call it a "tight but warm" bass-to-midrange tuning.

But, this is really just how I feel this morning. Get your system to be flat first, and then you can make good decisions on your own "house curve". Keep in mind the off-axis response of your speakers might fight you on certain EQ decisions.

I follow a similar strategy, I EQ to totally flat and then apply bass boost and cut some highs + apply a high-pass filter when needed

Currently this is what I use:

You can see the HPF there which I can turn on and off on demand
Note that I am not using shelf filters but simple bell filters since they don't keep the boost all the way but start to drop at some point (that you can tell exactly by looking at the EQ graph)

I believe there is no such thing as an ideal room curve - it is totally up to taste

You can try something like this: http://www.ohl.to/calculators/targetcurve.php but again.....just settle with what you find best for your taste!

 

[thewas]

The response at listening position depends also on the directivity of the loudspeaker, reverberation curve of the room and listening distance, so equalising to a fixed in-room target cannot work as also above the room transition frequency we mainly perceive the tonality of the direct sound.

More about it can be read at https://www.aes.org/e-lib/browse.cfm?elib=17839 (free access)
and https://www.audiosciencereview.com/...ut-room-curve-targets-room-eq-and-more.10950/

Thus I mainly equalise to flat anechoic direct sound above 200-300 Hz and below to a rising bass response that blends to the upper part with an approximate - 0.8 dB per octave slope.

 

[Sokel]

Just posted them here yesterday:

Late-night one

Loud listening.

 

[AetherDrive]

Hey guys, thank you for the responses! I meant to post the FR results that @Thomas_A asked for but I haven’t had a moment as shortly after posting this I got a job offer in another state which I have accepted. It’ll be a massive step in the right direction for my life, I just need to get all my stuff packed and ready to leave in Jan.

I should be able to post that FR chart on Monday as well as more considerately read over everyone’s posts, which again, thank you for everyone’s insights!

 

[D!sco]

My current flavor of house curve, -0.5dB/octave with a substantial sub-bass boost. High pass filters are 2nd order and are adjusted to the situation. Sometimes I leave the PEQ at 15000 off for a bit of lift at the top end.

 

[Le Concombre]

my latest making of goes like this

After a first steady state measurements made moving the umik microphone straight to the ceiling with REW recording RTA, I made preliminary set of corrections with 2 filters for R and 3 for L (those filters span from 61 to 156 Hz and are very broad ( Q of 0.27 to 0.67)) and got the attached after second set of measurements

Those results can be treated with success in automatic mode in REW (at initial stage it's too far off in my case) ; yeah but to which target(s) ??

from the exemple of Left channel, I consider that Toole fits my HF better than either Harman 4 dB (too high) or BK ( too low) , so my HF profile follows Toole.

As of LF, Toole (or BK, very close and used for ages (1970s) as house curves in many studios) is the safest bet. But why loose all that bumpy energy around 38 Hz ??

So, my number 1 curve is a Toole with the peak below 60 Hz only half shaved : gives weight to the lowest notes of a bass guitar without interfering with male voices nor classical instruments timbres
This being some recordings/masterings are too bass shy (because the monitors had a lot of room gain/bass lift in the mastering studio or intended Dre's headphones etc ??). So I also have a curve after Harman 4 for LF (keeping the parabolic profile of Toole for HF) and there too I only half shave the peaks around 38 hz

 

[kimmosto]

KS-1804 mk3, no room EQ

 

[Le Concombre]

Indeed the making of I described above is biased : the preliminary set of corrections with 2 filters for R and 3 for L (those filters span from 61 to 156 Hz and are very broad ( Q of 0.27 to 0.67)) was made with Toole in mind.
Going back to initial measurements, it appears that a straight line (at least above 450, below I have bass absorption issues to compensate for) fits nicely with a natural parabolic fall round 10 K
At the moment I'm thus enjoying the Harman with 4 db room gain (LF) but with natural parabolic fall ( I leave boosting HF above 10 K to match a straight line to masochists or people with speakers with bad directivity issues)

 

[Le Concombre]

A must watch :

thanks !!
I could not get satisfying results with step 2 (main correction). Maybe REW for Mac is buggy when it comes to special functions like inversion...Anyway step 2 allows step 3 and it's never been so easy/ well explained to work with RePhase and get no preringing to get filter 3. I did not venture in step 4.

The treasure is the Virtual Bass Array yielding filter 1 but I convolve the resulting filters (impulses) (f1 x f3 )in HQP on top (added on the same pipeline lines) of my .txt of filters obtained on steady state* (revised : new set of measurements with VBA + Harman 4 correction and I get a decent Toole) it's fast with amazing attacks, thanks to conter firing standing waves (filter 1) and precise impulses (filter 3)

*beware of ss for HF, as thewas mentioned, flat anechoic direct sound above 200-300 Hz is the thing though I only found real damaging discrepancies above 3K where steady state correction induced boosting a notch that appears real late when applying FDW

 

[ernestcarl]

A must watch :

thanks !!
I could not get satisfying results with step 2 (main correction). Maybe REW for Mac is buggy when it comes to special functions like inversion...Anyway step 2 allows step 3 and it's never been so easy/ well explained to work with RePhase and get no preringing to get filter 3. I did not venture in step 4.

The treasure is the Virtual Bass Array yielding filter 1 but I convolve the resulting filters (impulses) (f1 x f3 )in HQP on top (added on the same pipeline lines) of my .txt of filters obtained on steady state* (revised : new set of measurements with VBA + Harman 4 correction and I get a decent Toole) it's fast with amazing attacks, thanks to conter firing standing waves (filter 1) and precise impulses (filter 3)

*beware of ss for HF, as thewas mentioned, flat anechoic direct sound above 200-300 Hz is the thing though I only found real damaging discrepancies above 3K where steady state correction induced boosting a notch that appears real late when applying FDW

I'm skeptical of the need to have the convolution filter be so extremely finely detailed esp. in the upper half of the FR. It's basically trying to remove or cancel out the reflections very high up there (which we know is extremely position dependent) as well -- this interestingly creates quite an unnaturally, uneven decay pattern. *Examine closely what happens to the spectral decay and spectrograms of the filtered responses.

Now, to be clear, I'm not saying the very simplified filter example (orange curves) shown above is "best" for this particular room EQ example scenario. It's just something else to compare against...

 

[Le Concombre]

I'm skeptical of the need to have the convolution filter be so extremely finely detailed esp. in the upper half of the FR. It's basically trying to remove or cancel out the reflections very high up there (which we know is extremely position dependent) as well -- this interestingly creates quite an unnaturally, uneven decay pattern. *Examine closely what happens to the spectral decay and spectrograms of the filtered responses.

View attachment 268334

Now, to be clear, I'm not saying the very simplified filter example (orange curves) shown above is "best" for this particular room EQ example scenario. It's just something else to compare against...

I tried with windowing, I tried by smoothing... your explanation might be the right one ; anyway my filters for room correction remain made on smoothed steady state measurements.
And finally I only keep the filter obtained at the VBA stage. I have no observable preringing observable in REW but in the end, since I don't use the correction on which the observation is made, I hear pre ringing.

Anyway, you should definitely try the VBA step ; that cancellation of standing waves has not effect comparable to just lowering bass (in the end my H4 with no VBA and my Toole (H4 + VBA being very close to Toole I chose that target to refine the measured results of VBA +H4) are pretty close) and I have played long enough to know that all the details speed attacks multiplication of windows/clues on actual event don't come from the difference in steady state bass level

Note the notch I now leave as is at the top end, that does not exist in direct sound.

 

[ernestcarl]

tried with windowing, I tried by smoothing...

One has to use different settings so just try and see what works best. In general higher resolution should be used in the bass and less the higher up you go — essentially a sliding scale which REW does not do so one has to do this manually. I believe in the above I used 6 cycles in the upper half and 15 or more down low for the magnitude EQ. For excess phase extraction and inversion, I used 6 cycles — though, again, the amount of windowing required really varies per individual situation.

 

[ernestcarl]

I'm skeptical of the need to have the convolution filter be so extremely finely detailed esp. in the upper half of the FR. It's basically trying to remove or cancel out the reflections very high up there (which we know is extremely position dependent) as well -- this interestingly creates quite an unnaturally, uneven decay pattern. *Examine closely what happens to the spectral decay and spectrograms of the filtered responses.

View attachment 268334

Now, to be clear, I'm not saying the very simplified filter example (orange curves) shown above is "best" for this particular room EQ example scenario. It's just something else to compare against...

@Serkan actually has a number of videos on rePhase and REW and I just happened to have previously downloaded one of his example filters -- can't remember which it was. Looks like in this video, however -- just quickly checked -- his final correction filter appears to be windowed more so there is no super fine comb filtering effect as we saw above. Still, I find the remaining substantial level of "detail" for room EQ correction too much. But that's coming from me, I have a bias for MMM or wider spaced spatial averaging.

----

*edit: here's the original video with the mdat file where I got the measurements from:

Indeed, the filter I showed in that example was not actually his final correction filter. It's been a while since I saw any of his videos -- and they're not exactly short viewing -- so I've already lost track which measurements correspond with whatever. But, I'm going assume the one named "Lf1234" is the final version here.

Oh, and just to illustrate what I previously said, here are the spectral decay and spectrogram graphs of the filtered IRs:

 

[dualazmak]

Even though quite belated, I just noticed this thread for the first time.

Just for your reference, let me show my present best tuned FR curve;

You would please find here my latest system setup.

Also please refer to my post here for the safe and flexible on-the-fly high-Fq relative gain control for tweeters and super-tweeters adjusting to hearing level (age-dependent hearing decline) of the audience invited to my audio listening sessions.

 

[Le Concombre]

HELP PLEASE : question for REW users :

when I apply a FDW to a sweep measurement or a vector average, there's a drastic change in the HF profile that differs much (by an order of magnitude of 6 dB slope)
I thought it was a REW on Mac issue but I opened the mdat attached after the OCA video and windowing just gets rid of excess peaks and dips, does not change the profile/slope. I added my measurements to the OCA mdat and my measurements would still change HF profile with windowing : issue in the measurements settings ??

there might still be an issue/bug with step 2 of the video though : in his mdat the author uses a straight line while he presents a 2 slopes version in his video and I made things more complicated with the choice of Toole that ends parabolic : I ended with filters creating an upward parabolic ending... so, for whomever tries, maybe it's safer to use a straight line to avoid creation of symmetrical of intended

 

[ernestcarl]

HELP PLEASE : question for REW users :

when I apply a FDW to a sweep measurement or a vector average, there's a drastic change in the HF profile that differs much (by an order of magnitude of 6 dB slope)

Frequency dependent windowing discards the later arriving energy that goes beyond the number of cycles chosen. In a very dry room like mine, I do not see huge changes in the response. If a lot of the sound going into the mic comes much later in time or are reflected energy, it will be removed. Vector averaging at widely spaced measurements in a room where there's a lot of reflections/time difference in arrivals also causes cancellations or loss of energy -- kind of like a comb filtering energy loss which may be more pronounced the higher up in frequency you go -- but this does not necessarily mean you aren't hearing those filtered out late arrivals -- much of it is just spaced or spread out further away from the initial transient attack. For EQing the magnitude respone, RMS averaging is normally used... unless you require/want the time/phase information. Also, prior to vector averaging measurements from the same channel/speaker, make sure that the phases or IRs are aligned.

When FDW is applied to the rears (which are very distantly positioned) of my 7.1c setup, I see a bigger reduction in magnitude compared to the fronts and sides which are at a very nearfield distance. Because of the reflections, however, they sound much louder than what the steady state curves indicate. This is why when setting up the levels and broad peak PEQs/shelving, I also use my own ears and do not just rely on how the curves look on the graphs.