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

[tmtomh]

Last nights listen was quite good. Very satisfying bass. But there was a bit too much volume somewhere about the midrange, I'm thinking the 300-500 range. I'll have another listen in a while in case it was my allergy affected hearing.

I'll sort out an mdat upload later.


It does seem like a bit of a chicken or egg scenario, but with todays technology it makes sense to buy a speaker that is known to behave properly (or at least adequately for the money). DSP crossovers in particular are "perfect" in a way, as they don't suffer from issues relating to tolerance, age, and limited component selections (like the design might call for a 92ohm resistor but you can only get 90 or 95). Also you can do any slope with DSP and it doesn't mean more electrionic components (more pairs of capacitors and resistors).


How do you know they have good directivity though? By all accounts it makes a difference to have a smooth directivity index. But yes in the days of old passive speakers it does seem a limitation of the analogue crossover circuits in having accurate transition zones and not having 1 driver running hotter than the rest, etc.


I would go ahead and do the quasi-anechoic measurement method linked before. I did it with my old Aaron ATS-5 speakers as an experiment (only did the direct measurement and the close-up measurements though). The direct result looks surprisingly smooth, with only the tweeter running a little hot in the top end. I placed the speaker so that the top of the midrange driver was roughly 1.2m from floor and ceiling, and having the speaker in the middle of my 3m wide room meant it was 1.5m from the walls, and I put the mic at the same height 1m away.
View attachment 296026 View attachment 296025

I looked up your speakers and found an article listing the crossover points. Being a 5 way design, that's a lot of components that can go wrong! Eye-balling your curve, I reckon the bass drivers are working good, the 5" is working good, and the 3" is working good in the lower half of its range. But yer, something is not right around that 5kHz crossover point. Possibly a component has gone bad, or possibly that's just the way it was? I also wonder if the super tweeter isn't working at all, because it is meant to work above 10kHz but doesn't seem to be doing anything.
View attachment 296027

Maybe go ahead and start a thread on these speakers (if you haven't already)? Those speakers make for an interesting experience. I was kind of hoping my ATS-5 were a bit wonky in the response, which would give reason to actively amp them, like this bloke did with his old Yamaha speakers: Multi-Channel, Multi-Amplifier Audio System Using Software Crossover and Multichannel-DAC

Many thanks!

RE the super tweeter: you know, I hadn't even thought of that. The crossover to the super tweeter is indeed at 10kHz, but with my concern about the 5kHz issue it didn't even register that the upper treble decline starts right around 10kHz. I think I have the super tweeter's L-pad turned down pretty far, so I'll turn it back up a bit to see if that makes any difference. I would be a bit surprised if both super tweeters were non-functional, especially since my father never pushed these speakers, playing them at modest volume. But anything is possible.

And thank you for the link to the tutorial on quasi-anechoic measurements - much appreciated! As with every diagnostic option, the size and weight of these speakers makes everything more difficult - they are way too big and heavy for me to take outside or to raise up on a platform. However, in my case I'm really just looking for quasi-anechoic results above 2.5kHz, and both speakers appear to have functionally identical response (above the Schroeder frequency). So I can probably find out what I need to find out by just "walking" one speaker to the middle of the room and taking a quick on-axis measurement with a shorter than usual gate-time, since I only need measurement resolution down to about 2kHz.

I have to say, I never, ever would have bought these speakers, and the only reason I even went through the trouble of moving them into my room is because they were my father's. My own preferences tend towards simplicity and elegance of implementation, and while these guys certainly look elegant, they are the opposite of simple, and I don't think anyone would accuse their circuit design of being elegant.

 

[neRok]

So I can probably find out what I need to find out by just "walking" one speaker to the middle of the room and taking a quick on-axis measurement with a shorter than usual gate-time, since I only need measurement resolution down to about 2kHz.

Correct. Those speakers seem pretty tall, so whatever the distance is from 3" driver to the floor, get at least that far away from any walls and you should be good.

And just to make sure you don't miss the edit on my last post, here it is again;
if you don't want to go to the full effort of the quasi-anechoic method, at least do 2 measurements with the mic right up close to each 3" driver to ensure they are working the same (no faults with the drivers or crossovers). You can leave them where they are, because being close up, the direct sound will dominate the result. Perhaps do the same with the lower tweeter too (that crosses at 5kHz).

 

[tmtomh]

To do this would mean "fighting" the crossover electronics, because you will have to have push in so much power to overcome the electronics. If you have to push in +6dB, you've basically lost 6dB headroom on the system (or you risk clipping when running it loud).

Yes, I totally hear you on that. And in an indirect way it's one of the reasons I'm going to give up on these speakers if a couple more attempts at getting them right prove unsuccessful. I don't think headroom is an issue - per @Holdt 's comment above, the system seems to have enough headroom: I haven't gone above -18 on the MiniDSP's volume control, and that's only for playing quietly recorded stuff loud - my typical listening level is with the MiniDSP at -34 to -24.

But in principle I am with you - I don't like the idea of fighting a crossover issue by pumping more gain into that range via EQ - I don't think it will compromise headroom, but it does make me wonder what kind of distortion reading Amir would get at that frequency if he were testing this speaker. (Of course sending him a 5 foot tall, 2 foot wide, 120lb speaker is a non-starter for me. )

But that said, I've been using full-range correction in Dirac, and so it's already been doing some of that presumably anyway, and I haven't heard any horrible things at 5kHz that I'm aware of, so I'll give the PEQ option a try. If it sounds bad or is ineffective, I'll just try something else.

 

[tmtomh]

Correct. Those speakers seem pretty tall, so whatever the distance is from 3" driver to the floor, get at least that far away from any walls and you should be good.

And just to make sure you don't miss the edit on my last post, here it is again;
if you don't want to go to the full effort of the quasi-anechoic method, at least do 2 measurements with the mic right up close to each 3" driver to ensure they are working the same (no faults with the drivers or crossovers). You can leave them where they are, because being close up, the direct sound will dominate the result. Perhaps do the same with the lower tweeter too (that crosses at 5kHz).

Thanks!

Just so I'm sure I understand:

Mic right up close to the driver - that's just to see if each speaker's super tweeter is producing the same response as the other one, as a check to make sure they're functioning, yes? But the measurements won't actually provide accurate quasi-anechoic results, correct? For that I will still need to move the mic farther away from the speakers and take measurements from that distance?

 

[tmtomh]

You should then try to set your single peq boost in the minidsp before running dirac. That way you can already see if it worked as intended on the first measurement response in Dirac.

Oh, I hadn't thought of it - good idea. Makes sense: since I already have prior Dirac measurements that reveal that dip, why not try to correct it and then take a new measurement to see if it worked? Thanks!

 

[neRok]

that's just to see if each speaker's super tweeter is producing the same response as the other one, as a check to make sure they're functioning, yes?

Yer, just to make sure they're operating at the same level and rolling off the same. If 1 component of the crossover starts going bad, anything could happen (roll off higher or lower, different level, etc).

I haven't heard any horrible things at 5kHz that I'm aware of

True, it might not be an issue. I just chucked a random metal song from my playlist in to Audacity, and I'm not expert with how to use it, but it looks like the bulk of the songs sound is below 300Hz, and there's a fair bit up to 6kHz, but above that is a bit patchy and rolling off. It does seem pretty irrelevant above 10kHz (even bad speakers will have sound, so that's probably good enough).

 

[tmtomh]

Yer, just to make sure they're operating at the same level and rolling off the same. If 1 component of the crossover starts going bad, anything could happen (roll off higher or lower, different level, etc).


True, it might not be an issue. I just chucked a random metal song from my playlist in to Audacity, and I'm not expert with how to use it, but it looks like the bulk of the songs sound is below 300Hz, and there's a fair bit up to 6kHz, but above that is a bit patchy and rolling off. It does seem pretty irrelevant above 10kHz (even bad speakers will have sound, so that's probably good enough).
View attachment 296036

Thanks, and interesting! Yes, in general there is surprisingly little treble energy in most music.

Also, per your comments about the super tweeters, I'm off to bed so no time to arrange a new measurement, but I quickly created a 13.5kHz tone file with Audacity and just played it through both speakers. By removing the grilles and putting my hand over the tweeter and then the super tweeter, I was able to easily hear the tone coming from each separate driver, and both super tweeters were pumping out the tone just as loudly, if not more so, than the tweeters. So I think that's a pretty good indication that the super tweeters are working.

I also re-checked the rear L-pads for the super tweeters, and I had set them both to 7 o'clock aka their minimum, probably because I had quickly scanned that Dirac graph and misread the upper treble part of it. I have turned the super tweeter L-pads to 11 o'clock and will see if that has any impact when I take more measurements over the weekend.

The more I investigate this, the more I am leaning towards the issue being the crossover design and/or driver-integration problems caused by the replacement midrange. Given that the super tweeters are working, and the tweeters are working, and the speaker's measured response in the mid and highs are pretty much the same (and the soundstage is not suddenly shifting left or right at certain frequencies), I feel a crossover/midrange integration issue is more likely than the identical components of both speakers failing in identical ways to identical degrees.

However, with all that said, aging capacitors and other electronics in the crossover network certainly could be a contributing factor even if nothing is actually broken. And I lack the skill to refurbish the crossovers - and I'm just not willing to undertake the effort and expense that it would take to properly and safely get these things out of my basement and transport them to and from a qualified tech.

 

[Gringoaudio1]

You can manually EQ the entire frequency range. Room EQ software shouldn't be used to equalize frequencies above Schröder based on the in-room response.

Please provide citations.

 

[flipflop]

@Gringoaudio1

AES E-Library » The Measurement and Calibration of Sound Reproducing Systems

For decades, it has been widely accepted that a steady-state amplitude response measured with an omnidirectional microphone at the listening location in a room is an important indicator of how an audio system will sound. This paper examines both small and large venues, home theaters to cinemas...

www.aes.org

 

[Keith_W]

@Gringoaudio1

AES E-Library » The Measurement and Calibration of Sound Reproducing Systems

For decades, it has been widely accepted that a steady-state amplitude response measured with an omnidirectional microphone at the listening location in a room is an important indicator of how an audio system will sound. This paper examines both small and large venues, home theaters to cinemas...

www.aes.org

Thank you for that. I was most interested to read what Toole said in p.517 of your link:

It is essential to separate events above and below the transition/Schroeder frequency. Above it, at middle and high frequencies, constructive and destructive acoustical interference occurs when direct and reflected sounds combine at a microphone. If the frequency resolution of the measurement is sufficiently high—typically 1/3-octave or higher—the resulting peaks and dips can look alarming when seen in room curve. The tendency for a calibrating technician or automated equalization algorithm might be to attempt to smooth the curve. These are non-minimum-phase phenomena that are not correctable by minimum-phase equalization.

Although at present I follow Toole's approach and do not correct above Schroder, I was sent a link by a friend today where Bernt (author of Audiolense) quoted several studies that suggested the opposite: link.

I am open minded about whether my current approach is right or wrong. I am looking forward to a nice debate (no blood please!) on the merits of each approach.

 

[thewas]

Thank you for that. I was most interested to read what Toole said in p.517 of your link:



Although at present I follow Toole's approach and do not correct above Schroder, I was sent a link by a friend today where Bernt (author of Audiolense) quoted several studies that suggested the opposite: link.

I am open minded about whether my current approach is right or wrong. I am looking forward to a nice debate (no blood please!) on the merits of each approach.

I see him quoting there some papers of Olive which according to my memory don't show the anechoic response of the corrected and non-corrected loudspeaker as of course if the directivity is smooth and the target is chosen suitably you can end up linearising the anechoic response which is what Toole recommend (and works best in my experience after 20 years experiment with DRC).

 

[boxerfan88]

Here are my views:

• below Schroder, the room impacts the speaker frequency response significantly at those regions
• above Schroder, the room has lesser impact on the speaker frequency response (save for maybe omni's or dipoles)

Therefore, my preference is:

• EQ below Schroder to reduce the frequency response variability
• don't EQ above Schroder, and allow the speaker character to shine through

Except if:

• I don't like the speaker character, then go ahead to EQ to taste
• there is some issues caused by placement, then EQ to compensate.

 

[flipflop]

Although at present I follow Toole's approach and do not correct above Schroder, I was sent a link by a friend today where Bernt (author of Audiolense) quoted several studies that suggested the opposite: link.

He quotes 3 studies:
P. Hatziantoniou, J. Mourjopoulos, and J. Worley, "Subjective Assessments of Real-Time Room Dereverberation and Loudspeaker Equalization," Paper 6461, (2005 May.). doi:
S. Olive, J. Jackson, A. Devantier, and D. Hunt, "The Subjective and Objective Evaluation of Room Correction Products," Paper 7960, (2009 October.). doi:
S. Olive, T. Welti, and E. McMullin, "Listener Preferences for In-Room Loudspeaker and Headphone Target Responses," Paper 8994, (2013 October.). doi:

The first one is locked behind a paywall, so I can't comment on it.
The second is an evaluation of room correction products, which EQ below Schröder and may or may not EQ above it. A blog post about the study can be found here: https://seanolive.blogspot.com/2009/11/subjective-and-objective-evaluation-of.html
The third paper is also locked behind a paywall, but I happen to have a copy of it. I'm not even sure what this study is supposed to prove. It's about preferences in bass and treble levels for loudspeakers and headphones. It certainly doesn't support the notion that equalizing above Schröder based on the in-room response is a good idea.

It should also be mentioned that the last two studies are in the list of references of the paper I cited. One is not going to contradict Floyd Toole by citing Sean Olive.

 

[Keith_W]

He quotes 3 studies:
P. Hatziantoniou, J. Mourjopoulos, and J. Worley, "Subjective Assessments of Real-Time Room Dereverberation and Loudspeaker Equalization," Paper 6461, (2005 May.). doi:
S. Olive, J. Jackson, A. Devantier, and D. Hunt, "The Subjective and Objective Evaluation of Room Correction Products," Paper 7960, (2009 October.). doi:
S. Olive, T. Welti, and E. McMullin, "Listener Preferences for In-Room Loudspeaker and Headphone Target Responses," Paper 8994, (2013 October.). doi:

The first one is locked behind a paywall, so I can't comment on it.
The second is an evaluation of room correction products, which EQ below Schröder and may or may not EQ above it. A blog post about the study can be found here: https://seanolive.blogspot.com/2009/11/subjective-and-objective-evaluation-of.html
The third paper is also locked behind a paywall, but I happen to have a copy of it. I'm not even sure what this study is supposed to prove. It's about preferences in bass and treble levels for loudspeakers and headphones. It certainly doesn't support the notion that equalizing above Schröder based on the in-room response is a good idea.

It should also be mentioned that the last two studies are in the list of references of the paper I cited. One is not going to contradict Floyd Toole by citing Sean Olive.

I spent some time today trying to look up those references. I could not find any of those articles he cited freely available, but I did find this: Results for room acoustics equalisation based on smooth response. In it, he makes a mathematical definition of "Complex Smoothing" which I am unable to follow due to my rudimentary math skills. However, he does apply correction to the entire frequency range up to 15k, and a subjective preference score seems to indicate preference for the corrected version.

My own experiments with full range correction (with Acourate) produce unlistenable results. Subjectively, clarity seems to go down and there is a piercing resonant quality (I am guessing) at about 5-6kHz. The verification measurements look OK though - I just don't like the sound.

 

[OCA]

Harman curve was created for headphones not speakers but it works fine in rooms, too and REW auto EQ will yield descent results given:

1. No smoothing (practically 1/48 smoothing needs to be applied to remove notches from measurments) up to the room transition frequency around 200Hz where you listen to the room not the speaker. Up to 5dB boost for dips can also safely be used in that region in my experience.
2. No smoothing (1/48) between 200 and 500Hz (roughly where room effects end) but no dip boost allowed
3. Reflection free response above 500Hz (up to 20kHz) because in that region you correct for the speaker not the room, no dip boost as most dips here will be related to speaker crossover design. Anechoic chamber measurement is not practical for most users so a suitable windowing needs to be applied to the response to remove reflections i.e. if the distance of LP to the speaker is 3m, apply a right window size of 3 / 343 x 1000 = 8.7ms to the response which will only contain direct sound as no reflection can reach the LP faster than the direct sound. In practice, such tight windowing will not show any peaks to be corrected above 1-2kHz although there're exceptions in some cases.

 

[flipflop]

I spent some time today trying to look up those references. I could not find any of those articles he cited freely available, but I did find this: Results for room acoustics equalisation based on smooth response. In it, he makes a mathematical definition of "Complex Smoothing" which I am unable to follow due to my rudimentary math skills. However, he does apply correction to the entire frequency range up to 15k, and a subjective preference score seems to indicate preference for the corrected version.

The paper doesn't show the in-room and on-axis responses pre- and post-EQ and instead opts for exotic alternatives. There's a link to the "acoustic properties, detailed diagrams and auralizations for most the tested spaces", but it's dead now.
Table 3 shows the "Total percentage improvement [...] for the equalized signals, during the subjective listening tests." Instead of showing the improvement for each room, it shows what I presume is an average of all 6 rooms. It's possible that in certain rooms, the equalized signal was less preferred than the original.
The description of the methodology of the subjective listening tests also leaves me with the impression that they were not double-blind.

 

[tmtomh]

My own experiments with full range correction (with Acourate) produce unlistenable results. Subjectively, clarity seems to go down and there is a piercing resonant quality (I am guessing) at about 5-6kHz. The verification measurements look OK though - I just don't like the sound.

Subjectively I’ve not experienced a reduction in clarity with full range correction, but I have definitely experienced that piercing quality around 6k that you mention. Not with every combo of measurements and target curve, but often enough that it’s the main issue I have to contend with.

 

[Keith_W]

The paper doesn't show the in-room and on-axis responses pre- and post-EQ and instead opts for exotic alternatives. There's a link to the "acoustic properties, detailed diagrams and auralizations for most the tested spaces", but it's dead now.
Table 3 shows the "Total percentage improvement [...] for the equalized signals, during the subjective listening tests." Instead of showing the improvement for each room, it shows what I presume is an average of all 6 rooms. It's possible that in certain rooms, the equalized signal was less preferred than the original.
The description of the methodology of the subjective listening tests also leaves me with the impression that they were not double-blind.

Yes. Not only that, but they did not describe the listeners or the sample size. As a friend said, it's "wishy-washy" ... so not a great paper.

 

[Gringoaudio1]

Well I spent a few hours last night reading about ringing from boosting nulls in the EQ. I think from what I have gleaned I’m okay boosting minor speaker response driven FR dips but not room related bigger nulls. I’m EQing at the listening position with multiple readings in the vicinity of where I sit. I swear my system sounds way better with my EQ. Ha!
And thanks for the citations flipflop. I have some reading ahead of me.
And from the article I posted earlier cut narrow (highQ) and boost wide(lowQ) still applies. I see no reason why that isn’t a fundamental of EQing whether in production or at the room EQ level. Everything I read suggests that ringing can be minimized with shallow slope filters.

 

[OCA]

Well I spent a few hours last night reading about ringing from boosting nulls in the EQ. I think from what I have gleaned I’m okay boosting minor speaker response driven FR dips but not room related bigger nulls. I’m EQing at the listening position with multiple readings in the vicinity of where I sit. I swear my system sounds way better with my EQ. Ha!
And thanks for the citations flipflop. I have some reading ahead of me.
And from the article I posted earlier cut narrow (highQ) and boost wide(lowQ) still applies. I see no reason why that isn’t a fundamental of EQing whether in production or at the room EQ level. Everything I read suggests that ringing can be minimized with low slope filters.

Ringing, more specifically audible pre-echo will be introduced with sharp phase shifts below 500Hz. I can quantify sharp as larger than 45 degrees with a Q higher than 1. REW auto-eq generated filters are minimum phase and are very unlikely to generate phase shifts of that magnitude.

I know these limits because I create all possible filters and listen with these filters to find where ringing starts. In fact, ringing can be checked quite efficiently from step response.

I find it hard to understand that many audiophiles are so desperately seeking for post calibration results to even try a method. The most complicated filter takes less than an hour to generate. You cannot damage a speaker with a filter and if you don't like the results you switch your previous filter back on. Besides, flat frequency response is almost the least useful evidence of a well calibrated setup followed by phase response. I can get a FR flat with a PEQ by ear in minutes and it would most likely sound throttled/boxy.

I suggest people with passion for better sound out there to go and test the methods that sounds logical to them. Testing is learning.