EQ'ing to Harman curve doesn't give me pleasing results - why not?

[notsodeadlizard]

There is a simple practice-tested rule in engineering - the real semantics of the "may improve" expression in the general case does not coincide with the meaning of the word "improvement" at all.

 

[ZolaIII]

@neRok well my room is a little smaller (3x4 m) and it's far feald (3 m) similarly arranged setup of pair of mid size bookshelf's with fairly wide dispersion (Q3030i) and a 10“ sealed enclosure sub.
Had a grid power problem right after coffee and my amp went into power protection blocking mode (after power protection went on 3x in the row). Luckily it's resolved (resettled protection circuit) without any permanent demage to equipment at least.
How about puling the work table out a bit along with speakers and away from both back wall and corners (50 cm if you can)? It seems to me that you have space so you might. Point is to lower refractions from that wall. Treating it along with corners with 10 cm (4") rock wool absorber panels would also help noticeable for mids and highs so have that in mind. I ain't into R60, more to the ISO 3382-1. I don't prefer Harman low bass boost as I use ISO 226 2003.
For now try just with moving them out like I told you.

 

[Keith_W]

Like you, I used to correct my speakers to the Harman curve at the listening position. I complained about the awful sound in a thread somewhere in ASR (I forgot where). The advice I was given was - you should only correct below the Schroder (or up to 4x Schroder). Above 4x Schroder, the speaker should be corrected to its anechoic response. In my current system, I do nearfield measurements of the drivers and correct small bumps in the frequency response. I sort the crossover out and time align the drivers. After that, I leave it alone and only correct bass frequencies.

As audio wavelengths get shorter, the microphone becomes a model of what you hear. It is not what you actually hear.

 

[Matias]

Try using less filters on REW auto generated. Set like from 7 to 20 as Manual + None, that is, disabled, and let it generate with 6 filters only to see if less aggressive works better.

 

[ozzy9832001]

Honestly, I'd eq down the peaks first and check with each one. Leave the rest alone. Getting to flatter curve can cause issues from my testing. Once the peaks are under control, a shelf filter might sound better as the whole entire region until ~1k needs to be lowered at roughly the same dB. However, the speakers might sound better with more energy in the mids. That's more to personal tastes.

What your spectrogram shows me is you have a lot of ringing...everywhere. It's for sure a room issue. Speakers are in 2 corners which is giving significant gain to the system.

Can you pull the speakers away from the corner like a foot and then retest?

Have you tried setting up against the long wall instead of the short. My office is about the same size (slightly smaller on the long wall at only 11.5ft), and it's the best position because it keep me out of the corners and my speakers are roughly 14" from the front wall.

 

[sigbergaudio]

You're not supposed to force your speakers into the harman curve, but many well designed speakers will naturally be similar to that curve (with the exception of room nodes) in "normal" rooms.

Some simple guidance for the EQing (you've got similar advice already):
Don't EQ much above 2-300hz, and be careful between 100-300hz (rule of thumb, don't EQ more than 3dB). Below 100hz you can be a bit more agressive on the EQ if needed.
This will probably give you a decent result.

 

[flipflop]

Edit: This is the EQ that REW generated to the Harman target curve. I I think I set its auto-eq range up to 10k Hz.

Black: Kali IN-8 V2 on-axis response
Red: Kali IN-8 V2 on-axis response with your EQ settings
Ignore frequencies below the transition frequency.

Same graphs shown separately and with a 200 Hz tonality baseline added:

 

[tmtomh]

I'm still dialing in my new (to me) speakers, but it's been a few weeks and I think I'm either there or fairly close to what will be my final EQ. With that caveat said, my experience has been similar to some key points raised here:

• I've chosen a target curve based on the Harman preferences (understanding that there is no "Harman curve"), and as noted above, my speakers already generally adhere to that curve with their un-EQ'd in-room response. Most of what Dirac is doing is pulling down modal peaks and some smoothing.
• In the mids, my particular speakers have one area of elevation around 1.5-3kHz give or take, because of replacement drivers that are slightly more sensitive than the originals. Based on the measurements, it's broad (which sucks) but quite smooth (which is good), so it's easy to pull down, so I let Dirac do that and it seems to have worked very nicely.
• Above about 3kHz, I drove myself crazy with different downward target curves from there to 20kHz: -1.5dB at 20kHz, -3dB at 20k, -6dB at 20k. They all sounded too bright, too dull, or right in one area of that broad range but wrong in another. Finally I just said to hell with it and limited Dirac correction to a max frequency of about 3kHz. The speakers' raw in-room response shows a slight dip from about 4-7kHz and a slight bump from about 8-15kHz, but I left that alone except for a slight shelf filter of about -1dB from about 8k to 20k. So far, that's produced the best result to my (and my wife's) ears.

The speakers' response above 3kHz theoretically should sound better if I use Dirac there, since I assume it would be a bit more linear, but it just doesn't. If anyone has an idea about how to tweak the target curve in that area in a way that's more complex than the standard downward slope towards 20kHz, I'm happy to give it a try - but unless or until that happens, I think I'm sticking with my own minimalistic PEQ in the treble.

 

[sigbergaudio]

@tmtomh You should leave it alone as you have done now.

 

[tmtomh]

@tmtomh You should leave it alone as you have done now.

Thanks! That is my plan.

 

[fpitas]

Thanks! That is my plan.

I bet something is going on slightly off-axis. Best to EQ by ear in that case.

 

[tmtomh]

I bet something is going on slightly off-axis. Best to EQ by ear in that case.

Interesting. Dirac relies on 13 measurements including off-axis, but I would imagine that the way it averages/weights the measurements to come up with its filters - which are a bit of a compromise meant to produce a good result both on and off-axis - can't fully anticipate how the resulting filter will actually sound at every frequency to a human sitting at the main listening position.

I was glad that, in a fit of frustration, it occurred to me just pull Dirac's high-frequency correction limit down from 20k to 3k and generate a filter that way. But it's helpful to get some corroboration for that approach from folks here. Thanks!

 

[ozzy9832001]

Interesting. Dirac relies on 13 measurements including off-axis, but I would imagine that the way it averages/weights the measurements to come up with its filters - which are a bit of a compromise meant to produce a good result both on and off-axis - can't fully anticipate how the resulting filter will actually sound at every frequency to a human sitting at the main listening position.

I was glad that, in a fit of frustration, it occurred to me just pull Dirac's high-frequency correction limit down from 20k to 3k and generate a filter that way. But it's helpful to get some corroboration for that approach from folks here. Thanks!

Honestly, if you are in a single, static listening position then using all the measurement points may yield worse results than using only a couple. I found this to be the case. Center, Left back and Right back were all I needed. I'm fairly static when I listen to music, at least my head is. So, having all these extra measurements which may fall into null or peaks that are not at the MLP can skew the result. But I really have no idea how it uses the info.

At the end of the day, what sounds best by your ears is all that will matter. We all have different preferences which we consider good. Genre of music will have a huge impact as well.

Also, the question I always ask people is what is better sound? When messing with the EQ what does better sound sound like? How do you quantify it? If the music is missing something that's one thing, but if it's not, you're best leaving it alone and calling it a day.

 

[tmtomh]

Honestly, if you are in a single, static listening position then using all the measurement points may yield worse results than using only a couple. I found this to be the case. Center, Left back and Right back were all I needed. I'm fairly static when I listen to music, at least my head is. So, having all these extra measurements which may fall into null or peaks that are not at the MLP can skew the result. But I really have no idea how it uses the info.

At the end of the day, what sounds best by your ears is all that will matter. We all have different preferences which we consider good. Genre of music will have a huge impact as well.

Also, the question I always ask people is what is better sound? When messing with the EQ what does better sound sound like? How do you quantify it? If the music is missing something that's one thing, but if it's not, you're best leaving it alone and calling it a day.

I think you make very good points. As I read the Dirac manual, it provides three options: "very focused" with 9 measurements, "focused" with 13 measurements, and "wide" (or something like that) with 17 measurements. I chose the "focused"/13 measurement option simply because it was the middle option and I'd never used Dirac before. I've never actually checked into whether I could do even fewer than 9 measurements and then just skip to the "create filter" step. That might be interesting.

I also was under the impression that Dirac requires, or at least defaults to, a minimum of 9 measurements because it uses differences among the various measurements to create a quasi-anechoic model of the speaker's response vs the room effects. But of course I could be wrong about that.

As for what sounds better with EQ, I've always found it pretty obvious - typically tighter bass with fewer obvious peaks/resonances, and smoother mid/treble response with no objectionable peaking and hopefully no harshness.

However, I have managed to get generally pleasant, linear-sounding results with fairly crude EQ methods, for example a 10-band graphic EQ adjusted based on single-point readings at each octave with an SPL meter.

All that said, while I was waiting to receive my MiniDSP SHD unit after I got my latest pair of speakers, I did some quick and dirty EQ using the graphic EQ in my music playback software. The EQ immediately produced a better-sounding (to me, at least) result than no EQ. But after installing the SHD and creating some Dirac profiles, I disabled Dirac and re-enabled that graphic EQ setting in my music player app. And the sound was noticeably duller and less balanced than with the more sophisticated EQ in Dirac.

So based on what I've heard over the years, I lean towards EQ, especially high-quality digital EQ like with the SHD and Dirac, since the THD and noise penalty of such processing is negligible.

 

[neRok]

The speakers' response above 3kHz theoretically should sound better if I use Dirac there, since I assume it would be a bit more linear, but it just doesn't. If anyone has an idea about how to tweak the target curve in that area in a way that's more complex than the standard downward slope towards 20kHz, I'm happy to give it a try - but unless or until that happens, I think I'm sticking with my own minimalistic PEQ in the treble.

This is what I was wondering/rambling about in the OP. I don't know how REW/Dirac are technically handling the mic data, but they might be adding up (by ratios) 1 part direct with 1 part early reflection and 1 part late reflections and using that as the in room response, but what if I'm more sensitive to direct and/or early sounds and want 2:2:1 or 3:2:1 or even 1:1:0 for example? So I wonder if there is a way to EQ against a differently weighted "measurement"? I'm not sure if there is any validity to doing it this way, but I would be interested to try. I imagine this isn't possible with Dirac though, because that's a "closed box" afaik.

What your spectrogram shows me is you have a lot of ringing...everywhere. It's for sure a room issue. Speakers are in 2 corners which is giving significant gain to the system.

Can you pull the speakers away from the corner like a foot and then retest?

Yes my room is pretty plain. In the past, before I knocked up these stands, I had the speakers sitting on old loudspeakers (as stands), and I tested them in a bunch of different positions. I also looked at REW Room Sim. Both seemed to show that whilst in the corners had higher peakers, it also had higher (less worse) dips. Moving them out in either or both directions (in or forward) seemed to cause worse dips. But I will redo these tests because having them on the speakers might have been causing some effect to the sound?

Have you tried setting up against the long wall instead of the short.

I haven't, and I am under the impression that orientating in that direction is undesirable?

View attachment 294675Black: Kali IN-8 V2 on-axis response
Red: Kali IN-8 V2 on-axis response with your EQ settings
Ignore frequencies below the transition frequency.

Same graphs shown separately and with a 200 Hz tonality baseline added:
View attachment 294677

View attachment 294678

It looks pretty extreme when you put it like that! But when you look at my waterfall plots, you can see the reasoning behind the PEQ's below 3kHz. There is some "collateral damage" between the PEQ's, but I just figured that was the price you paid EQ'ing with such a method. But on the topic of too extreme of an EQ, I have seen other posters even on this forum that would take their in room response and subtract their target curve from it, and then use that difference as a FIR filter. I imagine if I did that, the -dB values wouldn't be too dissimilar to what is shown above. So what is the difference? Is that something I should be trying? (I probably should try it, just to know)

But you've prompted me to download the spin data, and I've loaded the estimated in room response into REW. It actually follows the Harman target quite closely from 130Hz to nearly 4kHz. And when I overlay those with my no_eq measurement, it actually doesn't look so bad. In fact it shows that the "manual" EQ I was doing with a PEQ at ~160Hz and ~280Hz was addressing the 2 major problems below 500Hz. The small rise between 400-600Hz is probably not worth addressing, but probably I need to handle the ~880Hz peak now too. Interesting.

What if anything can I do about the big dip at 1600Hz though? Could this be affected by speaker position? And are the peaks 2k and 4kHz worth addressing?

One other thing I just noticed on the waterfall is how the "axial length" room mode appears at 41Hz, and then multiples of, which includes my worst room mode at 164Hz. I wonder if I reduce the speakers performance at 41Hz, will that reduce the gain at the multiples? Probably not, but I will test anyway. Also if I then EQ the sub to handle that frequency, it will probably just bring the problem back anyway, as it is in roughly the same position with regards to the room length.

 

[ozzy9832001]

This is what I was wondering/rambling about in the OP. I don't know how REW/Dirac are technically handling the mic data, but they might be adding up (by ratios) 1 part direct with 1 part early reflection and 1 part late reflections and using that as the in room response, but what if I'm more sensitive to direct and/or early sounds and want 2:2:1 or 3:2:1 or even 1:1:0 for example? So I wonder if there is a way to EQ against a differently weighted "measurement"? I'm not sure if there is any validity to doing it this way, but I would be interested to try. I imagine this isn't possible with Dirac though, because that's a "closed box" afaik.

Yes my room is pretty plain. In the past, before I knocked up these stands, I had the speakers sitting on old loudspeakers (as stands), and I tested them in a bunch of different positions. I also looked at REW Room Sim. Both seemed to show that whilst in the corners had higher peakers, it also had higher (less worse) dips. Moving them out in either or both directions (in or forward) seemed to cause worse dips. But I will redo these tests because having them on the speakers might have been causing some effect to the sound?

I haven't, and I am under the impression that orientating in that direction is undesirable?

It looks pretty extreme when you put it like that! But when you look at my waterfall plots, you can see the reasoning behind the PEQ's below 3kHz. There is some "collateral damage" between the PEQ's, but I just figured that was the price you paid EQ'ing with such a method. But on the topic of too extreme of an EQ, I have seen other posters even on this forum that would take their in room response and subtract their target curve from it, and then use that difference as a FIR filter. I imagine if I did that, the -dB values wouldn't be too dissimilar to what is shown above. So what is the difference? Is that something I should be trying? (I probably should try it, just to know)

But you've prompted me to download the spin data, and I've loaded the estimated in room response into REW. It actually follows the Harman target quite closely from 130Hz to nearly 4kHz. And when I overlay those with my no_eq measurement, it actually doesn't look so bad. In fact it shows that the "manual" EQ I was doing with a PEQ at ~160Hz and ~280Hz was addressing the 2 major problems below 500Hz. The small rise between 400-600Hz is probably not worth addressing, but probably I need to handle the ~880Hz peak now too. Interesting.
View attachment 294753
What if anything can I do about the big dip at 1600Hz though? Could this be affected by speaker position? And are the peaks 2k and 4kHz worth addressing?

One other thing I just noticed on the waterfall is how the "axial length" room mode appears at 41Hz, and then multiples of, which includes my worst room mode at 164Hz. I wonder if I reduce the speakers performance at 41Hz, will that reduce the gain at the multiples? Probably not, but I will test anyway. Also if I then EQ the sub to handle that frequency, it will probably just bring the problem back anyway, as it is in roughly the same position with regards to the room length.

Speaker orientation is largely room dependent. In my office, setting up along the long wall is the better sounding, even though the distance from the 2 side walls is very asymmetrical. Along my short wall, I end up in a null at 80hz that I can't escape. I would try multiple positions and speaker/listening height to see which gives you the best overall response. General rule is to stay away from the middle of the room, but even that really depends. I always try to work in 5th for the room.

I don't see a dip at 160 on any of your measurements.

As for the waterfall, it's modal issues all over the place up to about 200hz. They are all stacking on top of or near one another. This is common in small spaces. (L)40, 80, 120, 160; (W)58, 116, 174; (H)72, 144. Those are your harmonics for your primary axial modes. Further, you have tangential and oblique modes which are also compounded. I imagine if you ran an RTA while listening to some music at your desired volume levels, you would see further spikes as the room fills with sound and the room modes all converge on one another. Oblique are the least impactful of the 3, but also the easiest to excite.

Those are the modes for my room. In this situation, treatment is a valid option to help quell the effects of them. You can see how they start to stack on top of one another.

Reducing the gain at 40hz will have an impact at 40hz and some at 80hz and probably not much at 120 or 160. Each frequency needs to be controlled separately because there are multiple ones compounding the issue.

As for the stuff over 500hz...I wouldn't touch. I'm a fan of leaving it only or using shelf filter to manipulate the entire stream. It has the peaks/valleys for a reason based on the speaker and its interaction with the room.

Can you post your spectrogram? Burst decay, either 1/3 or 1/6 doesn't matter, 30 second span, linear peak %?

Sorry, now I'm rambling.

 

[flipflop]

But when you look at my waterfall plots, you can see the reasoning behind the PEQ's below 3kHz. There is some "collateral damage" between the PEQ's, but I just figured that was the price you paid EQ'ing with such a method.

Exchanging on-axis frequency response linearity for a reduction in CSD is not a trade-off worth making.

But on the topic of too extreme of an EQ, I have seen other posters even on this forum that would take their in room response and subtract their target curve from it, and then use that difference as a FIR filter. I imagine if I did that, the -dB values wouldn't be too dissimilar to what is shown above. So what is the difference?

FIR EQ can make the frequency response hug the target more accurately and can also correct phase problems. Perceptually, there's not much to be gained.

Is that something I should be trying?

Be my guest, but if you think it will solve the problems caused by equalizing the in-room response to a target curve above the transition frequency, you will be wasting your time.

What if anything can I do about the big dip at 1600Hz though?

Improve your room acoustics or simply ignore it.

Could this be affected by speaker position?

Yes.

And are the peaks 2k and 4kHz worth addressing?

Not through EQ.

 

[Dal1as]

This is a good example of why room acoustics matter, knowing a speakers anechoic frequency response both on axis and off axis can be beneficial.

Also why even off axis response makes eqing a speaker easier.

You need to take into account 1st reflections from the ceiling, floor, walls, even your desktop.

A speaker with bad off axis response will tend to enhance certain frequencies more on axis because you are listening to the room now not just the speaker anechoicly.

 

[Gringoaudio1]

In summary this article says that for doing an EQ cut with narrow Q and boost with wide Q. I felt it gave me permission to boost as many people say to cut only.

EQ Tutorial - Cut Narrow And Boost Wide Explained

EQ Tutorial - Cut narrow and boost wide explained

www.loopmasters.com

 

[sigbergaudio]

In summary this article says that for doing an EQ cut with narrow Q and boost with wide Q. I felt it gave me permission to boost as many people say to cut only.

EQ Tutorial - Cut Narrow And Boost Wide Explained

EQ Tutorial - Cut narrow and boost wide explained

www.loopmasters.com

You can boost if you understand what you're doing, have enough headroom and don't try to boost nulls that can't be boosted.

EDIT: Also note that the link you provided is about EQing a mix not the room. So it's about music production. Quite different scenario.