X-curve compensation re-EQ

[SME]

It time for me to update things here.  Following on the "hidden resonances" insight I had, I developed a novel and vastly superior room and speaker EQ algorithm.  This took a long time and is still a work in progress.  However, I recently reached a substantial milestone, having implemented and fine-tuned the latest iteration on all 5.1 channels.  I had to develop and code some custom algorithms to get to the point I'm at now.  I've been having a lot of fun, doing real, serious math for the first time in a while. 

Previously, I was relying on estimates of first arrival response using FDW applied to measurements at different spatial locations, as inspired by Dr. Toole and Harman's work on polar, anechoic chamber measurements of speakers.  My new approach is totally different and all but abandons use of FDW entirely.  It's not that FDW was a poor choice of approach.  It actually worked very well for me compared to other room EQ systems I've heard.  It offered a theoretically compelling solution to the problem of choosing an optimal target curve to fit frequency response to.

However, my new method sounds so much better.  My recent experiences have totally changed the way I understand audio in terms of frequency response, whether in-room or "on the wire".  My new approach is based on a completely new theory of audio perception, one which I developed to try to reconcile my accumulated knowledge and experience with sound.  There's a lot of stuff that seemed weird before that now makes a lot more sense.

In any case, I spent some time today reviewing a number of movies to find out how they sound now.  I watched several scenes from both "Wonder Woman" and "Star Trek", both movies that I attempted to "re-master" and posted tentative re-EQ for.  Needless to say, I didn't even bother trying to apply any re-EQ, and both films sounded *excellent*.  That's not to say that they are perfect.  In fact, I can definitely still hear the increased emphasis on bass and/or treble in most of these mixes compared to most music.  However, in my recent viewings, these imbalances were *far less objectionable*.

What was happening before is that the broadband accentuation of bass and/or treble on the tracks was accentuating nasty resonances in my playback system in those ranges.  It was not the overall level of low frequency sound in the dialog but rather the finer-scale resonances in the low frequencies that were causing the upward masking / mud.  Likewise, much of the ear discomfort and downward masking apparently caused by the excess of high frequencies was actually caused by finer-scale resonances there as well.  Because my new EQ approach minimizes those degrading resonances, there is nothing for the broadband bass and treble boosts to accentuate.  This mostly eliminates the masking problems I was having, and I can hear the mid-range quite clearly throughout.

As such, I have much less personal motivation to re-master cinema tracks in the first place.  In a way, that's unfortunate because I still think that good re-EQ would help the tracks to sound better on a wider variety of systems.  At the same time, I can see the extra bass and/or treble as being preferred, for those whose systems can rendering it cleanly enough to not kill the mid-range.

There is also the philosophical question of director's intent.  Chances are very high that my presentation of e.g. "Wonder Woman" sounded better in my home theater than it did on the dub stage, but the director has never heard my system.  Would she approve?  I mean, her intent for a highly bass-focused and physical sound presentation is discussed in the articles linked above.  IIRC, her mixer also talks about getting the most bass out, given the constraints of a 120 dB SPL (informal?) budget.  My system as configured probably blew way past that.  I'd have to either measure (using a better mic than I own) or analyze the soundtrack data to determine where I peaked, but I don't doubt I pushed near 130 dB SPL.  If the director had 130 dB SPL instead of 120 dB SPL to work with, would she have used it?  Would the presentation be closer to my own?  I'd certainly like to think so.

Anyway, I will be viewing a lot of movies over the next few weeks / months and may decide to try to re-master stuff at that point, but it's definitely not a priority for me anymore.  It probably makes more sense for me to focus on developing my speaker/room EQ tech further because that made all the difference in the world.

[maxmercy]

Are these enclosure/horn/driver or room resonances you are correcting?

JSS

[SME]

2 hours ago, maxmercy said:

Are these enclosure/horn/driver or room resonances you are correcting?

Yes and more, including SBIR effects.  The method I use doesn't really care about what causes the features it sees.  Of course it can only correct problems that are linear and which are not strongly dependent on room location, even though it still tries.  Bass problems do tend to be more local than for mid and high frequencies, but they are nowhere near as local as one would expect by looking at standard frequency response measurements.  The ears and brain are remarkably good at listening "through the room" for the sound produced by the actual speaker(s)/sub(s), and this appears to be true even for very low frequencies, well down into the sub-bass range.

My process also avoids creating new resonances, which appears to be a major problem with most if not all other room EQ systems, including earlier iterations of my own.  Fundamentally, an *optimal* in-room frequency response at any location *will not be smooth or flat* unless the room is anechoic, and an anechoic room sounds bad.  Every room EQ system I know of tries to "correct" in-room frequency response and/or impulse response in some manner, which I have found is completely misguided.  It ignores the fact that listeners hear through most localized acoustic effects from diffraction, reflections, and so on.  My experience suggests that *this is true for the full range of frequencies*, including low frequencies where modal effects may be strong.

A floor bounce alone will contribute substantial ripple to the upper bass / low mid-range part of a single in-room frequency response measurement, and attempting to correct this, even using short-time and/or frequency-dependent windows, will just add new audible resonances to the speakers' sound.  Performing correction based on multiple spatial measurements can reduce the negative effects, but the particular choice of locations still biases the aggregate in some way or another and leads to creation of new resonances.  OTOH, the floor boundary (and any others that are nearby) alters the acoustic impedance adjacent to the speaker, affecting its acoustic power output sensitivity/efficiency vs. frequency.  This has a global impact on the sound produced, and *precise* correction of these effects is beneficial.  Modal resonances have similar acoustic loading effects, and again the idea is to correct their impact on the speaker/sub output, *not* the effect on response at any particular location.

[maxmercy]

18 hours ago, SME said:

A floor bounce alone will contribute substantial ripple to the upper bass / low mid-range part of a single in-room frequency response measurement, and attempting to correct this, even using short-time and/or frequency-dependent windows, will just add new audible resonances to the speakers' sound.

I can see this, by correcting response at one location, you create problems and ringing at others.  

 

18 hours ago, SME said:

OTOH, the floor boundary (and any others that are nearby) alters the acoustic impedance adjacent to the speaker, affecting its acoustic power output sensitivity/efficiency vs. frequency.  This has a global impact on the sound produced, and *precise* correction of these effects is beneficial. 

How can 'precise' correction of a reflection be 'corrected' for many locations?  The peaks and dips will occur at different freqs depending on location from the speaker..

JSS

[Ricci]

On ‎6‎/‎30‎/‎2018 at 10:31 PM, maxmercy said:

I can see this, by correcting response at one location, you create problems and ringing at others.  

 

How can 'precise' correction of a reflection be 'corrected' for many locations?  The peaks and dips will occur at different freqs depending on location from the speaker..

JSS

Good questions. Your first comment is why I stepped back from excessive amounts of EQ and trying for "perfect" graphs, even ones taken over large areas and averaged.  They usually sound very bad despite looking great and allowing forum bragging rights. Notch type high "Q" equalization is especially bad to my ears usually.

[SME]

On 6/30/2018 at 8:31 PM, maxmercy said:

On 6/29/2018 at 6:36 PM, SME said:

A floor bounce alone will contribute substantial ripple to the upper bass / low mid-range part of a single in-room frequency response measurement, and attempting to correct this, even using short-time and/or frequency-dependent windows, will just add new audible resonances to the speakers' sound.

I can see this, by correcting response at one location, you create problems and ringing at others. 

Yes, but this is true *not just at other locations* but also at the location where the single measurement was taken.  The ear and brain use information from reflections to hear through most acoustic effects that are particular to a single location.

On 6/30/2018 at 8:31 PM, maxmercy said:

On 6/29/2018 at 6:36 PM, SME said:

OTOH, the floor boundary (and any others that are nearby) alters the acoustic impedance adjacent to the speaker, affecting its acoustic power output sensitivity/efficiency vs. frequency.  This has a global impact on the sound produced, and *precise* correction of these effects is beneficial.

How can 'precise' correction of a reflection be 'corrected' for many locations?  The peaks and dips will occur at different freqs depending on location from the speaker..

A *reflection* cannot be corrected for multiple locations using DSP, and often one should probably not try to "correct" reflections because they actually facilitate the hearing process.  (Note: some reflections may still be degrading for some frequencies.)  To oversimplify just a little:  What we wish to correct is the effect of one or more *boundaries* (among the many other things) on the speaker's *total acoustic power output* response, which does not depend on room location.

On 7/2/2018 at 11:05 AM, Ricci said:

Good questions. Your first comment is why I stepped back from excessive amounts of EQ and trying for "perfect" graphs, even ones taken over large areas and averaged.  They usually sound very bad despite looking great and allowing forum bragging rights. Notch type high "Q" equalization is especially bad to my ears usually.

Yes.  Your anecdote was puzzling to me for a while, but not anymore.  It makes perfect sense now.  Of course that doesn't mean that EQ (even high "Q") can't improve sound.  Rather, the problem is that we misinterpret smoothed FR graphs, which don't really show us what listeners will hear.

[maxmercy]

14 hours ago, SME said:

What we wish to correct is the effect of one or more *boundaries* (among the many other things) on the speaker's *total acoustic power output* response, which does not depend on room location.

I do  not follow...  How can one correct something that is inherently designed into a speaker (power response), if it does not depend on speaker location?  Or are you talking about correcting power response depending on the speaker's location in the room?  Is this why incredible amounts of headroom are needed?  Is there not a simpler way to do this by changing the possible layout and treatments in the room, or is this correction geared towards getting as much as possible from a current configuration if placement and treatment options are limited?

JSS 

[SME]

On 7/7/2018 at 5:42 PM, maxmercy said:

On 7/7/2018 at 2:37 AM, SME said:

What we wish to correct is the effect of one or more *boundaries* (among the many other things) on the speaker's *total acoustic power output* response, which does not depend on room location.

I do  not follow...  How can one correct something that is inherently designed into a speaker (power response), if it does not depend on speaker location?  Or are you talking about correcting power response depending on the speaker's location in the room?  Is this why incredible amounts of headroom are needed?

Allow me to clarify my statement above: The speaker's *total acoustic power output* response does not depend on the location of the listener.  It does, however, depend on the location of the speaker.

With that said, there's no reason not to correct power response issues that occur in the native (anechoic) response of the speaker.  In fact when analyzing in-room measurements, it's not really practical to distinguish issues that depend on the speaker location vs. those that don't.  Very few speakers have ideal native (anechoic) power response either.  Even if the mid/woofer drivers measure very cleanly when on an I.B. and the cabinet doesn't have any panel resonances, the cabinet shape still contributes variations.

On 7/7/2018 at 5:42 PM, maxmercy said:

Is there not a simpler way to do this by changing the possible layout and treatments in the room, or is this correction geared towards getting as much as possible from a current configuration if placement and treatment options are limited?

In practice, placement and treatment options are almost always limited and serve as only partial solutions.  Often these options also involve compromises.  For example, most speakers sound their best in a room when aimed a certain direction, but this makes a flush-mounted installation difficult, especially with overheads and surrounds.  Also, any absorption removes valuable reflected sound energy as a side effect.  (This assumes small rooms where decay time reduction is rarely necessary, except at the modal resonances.)

Even in the most ideal circumstances, there is likely to be benefit from DSP if it's done well.  That leads to an interesting question:  To what extent is it possible to work-around acoustical problems using DSP?  The answer obviously depends on the capabilities of the DSP and quality of the algorithms used to compute the filters.  So what if one uses the best possible DSP capabilities and algorithms?  Well, no one can really answer that question because the best possible algorithms probably haven't been invented yet, and the best available algorithms may not be very good.

Conventional wisdom says that DSP cannot fix most acoustical problems and can only optimize sound at one location or perhaps compromise for handful of locations.  This reasoning is entirely valid from the view that the goal of DSP correction is to fix "problems" in the in-room response measurement.  The name of the company that produces the Dirac Live software refers directly to the goal of most room EQ system: to achieve an in-room impulse response measurement that looks more like Dirac Delta, which corresponds to a perfectly flat frequency response.

But what if all this conventional wisdom is wrong?  Empirical evidence suggests that anechoic chambers make terrible listening rooms, yet they are the closest to achieving an ideal Dirac Delta.  OTOH, 99.9% of the listening we do in real life is in rooms with significant reflections.  Maybe the reflections don't harm sound quality at all.  Maybe the real "problem" with reflections is that they confound our ability to measure and correct the speaker itself.  Or that they help reveal the less-than-ideal power response of the many speakers that otherwise look great when measured only on-axis.

Anyway, my recent experience suggests that DSP (done well) *can mostly work around* the kinds of acoustical problems that affect subjective sound quality and not just for a single listener location.  The filters do usually require extra headroom to implement.  Therefore, changing speaker placement or installing absorptive treatments may be beneficial in conjunction with DSP to reduce the amount of boost required.  Some boosts will still likely be required to overcome limitations of the speaker itself.

Edit: The text editor widget glitched and wouldn't let me add another paragraph to my response!

One caveat to add to all of the above is that a human listener can only ever *estimate* the power response of the source from the available information.  For the most part, these estimates can be remarkably accurate, perhaps excepting particularly pathological rooms or listener locations very close to untreated boundaries.  However, the accuracy of the estimates does deteriorate for bass in small rooms where the sound field becomes highly structured.  As such, the conventional wisdom that one can "only optimize sound at one location or perhaps compromise for a handful of locations" does apply to bass to an extent, particularly below roughly 150 Hz depending on room size.  However, this is not nearly as bad as one would expect by looking at in-room FR measurements.  The subjective variance is still much less than what is observed directly in the measurements.

Several strategies may be used to reduce subjective variation of bass.  One example we're all familiar with is to place subs in multiple room locations.  I suspect this may be beneficial even if the placements are not optimized, e.g., to cancel modes.  Of course, independent filters on multiple subs has the potential to do even better.  When I did this before, I was optimizing in-room frequency response, which didn't sound nearly as good as I'd hoped.  I expect that optimizing for a superior objective will deliver much better subjective results.

Edited July 10, 2018 by SME

[3ll3d00d]

fwiw this is basically the way roomperfect is marketed (albeit with a healthy dollop of audiophile voodoo mixed into the language used)

[SME]

9 hours ago, 3ll3d00d said:

fwiw this is basically the way roomperfect is marketed (albeit with a healthy dollop of audiophile voodoo mixed into the language used)

Thanks for the reference.  Since I had not heard of roomperfect, I decided to visit their web site to try to learn more about their product and the marketing language.  The main web page was sparse, but I found a bit more info on this article:

http://lyngdorf.com/news-what-is-room-correction/

To their credit, they seem to dedicate effort to improving in-room speaker power response.  However, as best as I can tell from their description, their methods are nowhere near sufficient to obtain an unbiased measurement of in-room power.  There is another, potentially larger problem:

Quote

[...] We realized we should not have all speakers calibrated to the same target curve, because doing this means all speakers would sound nearly identical. [...] All loudspeakers sound different, and you choose a given model because you love its performance and looks. The type of loudspeaker, its design, and the choice of components will create a specific and desirable product performance. [...]

Here they seem to be implying that one of their goals is to preserve the (presumably desirable) unique sound signature of the customer's speakers.  This tails in with the marketing of the product as a "room correction" product rather than a "speaker correction" or "EQ optimization" product.  So how exactly do they achieve this goal?  How do they distinguish between the speaker's "specific and desirable product performance"  and the room's degrading influence and only correct for the latter?  They don't explain how, and I very much doubt they have any way of making this distinction in the first place.

I believe the industry of "room correction" products has a kind dirty secret.  To the extent they work at all, it's mostly about speaker correction not room correction.  I can offer two possible reasons why the technology is marketed this way.  First is simply ignorance.  Room effects cause the vast majority of variation within in-room frequency response measurements, leading naive engineers to erroneously conclude that the room is overwhelmingly to blame for poor sound quality.  Second, "room correction" probably sells much better than "speaker correction".  Most audiophiles don't want to be told that their speakers (possibly costing 5 or 6 figures) are flawed.  It's much easier to point to the huge variations in-room response measurements (+/-20 dB !!!) and sell people on the idea that *their room that is holding back their speakers*.

I realize I have emphasized room-dependent problems in my discussion above, but my focus is more general.  IMO, the best description for my approach is "in-room speaker EQ optimization".  It is intended to take the place of two activities which are typically performed separately: voicing and crossover design (usually performed using anechoic measurements) and so-called "room correction".  I personally could care less about preserving the "unique sound signature" of particular speakers or other "audiophile" products.  I just want the best sound possible from my system.

[3ll3d00d]

Yes that is part of the audiophile style marketing they use. You will also find not many measurements are available to verify what it is actually doing and how well it performs. FWIW I think https://patents.google.com/patent/US8094826?oq=8094826 is their patent, it seems pretty generic to me though.

To be fair I have listened to their high end kit and it is good so there some substance there, it isn't just audiophile blather.

 

 

[SME]

15 hours ago, 3ll3d00d said:

Yes that is part of the audiophile style marketing they use. You will also find not many measurements are available to verify what it is actually doing and how well it performs. FWIW I think https://patents.google.com/patent/US8094826?oq=8094826 is their patent, it seems pretty generic to me though.

To be fair I have listened to their high end kit and it is good so there some substance there, it isn't just audiophile blather.

I would say the patent gives a lot of details, even if it does not give a lot of specifics.  It's way more informative than the marketing literature is, at least.  The gist of the method(s) described by the patent is as follows:

Fundamentally, EQ is optimized so that a measurement or averaged cluster of measurements at/around the listening position are made flat.  However, the measurement/average is potentially smoothed and filtered before the correction is computed, and the correction is subject to constraints.

No additional information is given about the smoothing.  That's too bad because the particulars of the smoothing potentially has a big impact on the quality of the result.  They do describe performing measurements at 1/12th octave resolution, which is pretty "meh" as far as these things go.  That's the good news.  The bad news is that the measurements rely on pure tone sine waves, so the results will have a lot of uncertainty when used to estimate the shape of the whole spectrum.

Before the correction, the response is pre-conditioned using filters.  The filters are derived from idealized models of normal and expected in-room frequency response and/or power response.  They essentially take the place of a non-flat EQ target curve by reshaping the listening position response (or average) before calculation of the EQ parameters.  The models discussed in the patent include: a high frequency directivity roll-off (expected more in power response than at listening position), low frequency room gain, and low frequency high pass / roll-off.  The parameters for the pre-conditioning filters may be calculated using the measurement itself (HF and LF roll-off models) or may be based on foreknowledge of certain characteristics (bass room gain model).  The end result is very similar to developing and applying a target curve that is customized to the room and speakers and possibly to listener preference.

The constraints are derived by taking additional measurements around the room and averaging them.  This average is intended to be a power response estimate.  (Based on my experience, the quality of this estimate is probably very crude.)   The power average is also pre-conditioned with filters, just like the listening position measurement (or average) was.  The pre-conditioned power average is then inverted, and the inverse is used to develop upper and lower bounds on the EQ.  The idea here is to prevent the EQ from doing anything too extreme (in either direction) to the power response.

===

There are many likely benefits to their approach.  The pre-conditioning filters effectively customize the target curve for the room and speaker.  I'm a big critic of EQing to a one-size-fits-all target curve, so this is a big plus to me.  Their models are probably over-simplified, but I would guess that they make a big improvement over one-size-fits-all.  My guess is that they stumbled upon this approach when trying to figure out how to re-EQ speakers close to walls.

The use of a power response estimate to develop EQ constraints is also a big plus.  This is a solution for the "hidden resonances" problem I described above.  However, I don't believe their's is a very good solution for a few reasons.

The worst part is that their measurements rely on pure sine wave test tones.  In anechoic measurements, pure tones tend to be OK because there are no acoustic effects (except involving the speaker itself).  The impulse responses aren't very long except for high Q low frequency resonances, which are fairly rare in competent designed speakers.  In-room, it's a totally different story.  Acoustic effects will effectively contribute a lot of uncertainty to the measured values.  It's because a measurement at a narrow frequency is a poor statistical guess about what FR is doing in a fuzzy region around that frequency, which is the kind of information that's needed to do the correction.

Their tech note indicated that pink noise was rejected as a test signal because it didn't "reach deep enough into the impulse response".  I'm pretty sure that's wrong.  The pure tones certainly engage the late parts of the impulse response but they don't reveal them to the measurement system any better than the pink noise does.  The most likely reason they chose pure tones over pink noise was because the signal-to-noise ratio was better.  Unfortunately for them, the "insight-to-signal" ratio using pure tones is not so good.  Moving to REW-style log sine sweeps and applying appropriate smoothing would probably lead to big improvement in the performance of their system.

I would argue that they are also worrying too much about the listening position measurements.  Practically everybody assumes that in-room frequency response will tell them something about what they'll hear in that position, but the reality is that things are a lot more complicated than that.  The information is there, but the primitive analytical techniques in widespread use don't do an especially good job of recovering it.  Power response is a major key to the puzzle, but it's not obvious how to determine speaker power response using in-room measurements.  As I said, the approach taken in this example is quite crude, and would still be crude even if the measurement methods were improved.

To follow-up on my comments about room vs. speaker correction:  The best "room correction", by a long shot, is the space between your ears.  Listeners are extraordinarily well-adapted to listening in environments with early reflections, and experimental evidence suggests that listeners *hear better* in the presence of early reflections.  As such, the goal is not to correct the room but to correct the speaker for the room and let the brain do its thing.  The big problem is that no one has a particularly good perceptual model to apply to in-room IR/FR measurements to assess what listeners will actually hear.  Almost every product relies on EQing in-room response to some kind of simplified target, and each such product adds its own twist to make it unique to make it not actually suck.

The use of a power response estimate to constrain the EQ is very good idea, but the methods described in the patent for doing so are rather obvious and primitive, IMO.  Still, I don't doubt that it sounds pretty good, especially compared to other room EQ systems.  It's possible that they have improved their analytical methods some beyond the patent.  However, the tech note suggests that they are still using the primitive pure tone measurements, and I believe that holds it back considerably.

[maxmercy]

11 hours ago, SME said:

The best "room correction", by a long shot, is the space between your ears.  Listeners are extraordinarily well-adapted to listening in environments with early reflections, and experimental evidence suggests that listeners *hear better* in the presence of early reflections.  As such, the goal is not to correct the room but to correct the speaker for the room and let the brain do its thing.  The big problem is that no one has a particularly good perceptual model to apply to in-room IR/FR measurements to assess what listeners will actually hear.  Almost every product relies on EQing in-room response to some kind of simplified target, and each such product adds its own twist to make it unique to make it not actually suck.

 

So what would you do if you had what most people on forums have available: limited budgets, limited placement options, some room treatment, REW measurement capability and limited parametric/IIR correction capability?  

Would you mainly focus on minimum phase problems in the LF and largely ignore MF/HF, or how would you go about optimizing a system with limitations like the above, given what you know now?

I ask not only because that's the way my system is set up, but many others with miniDSP or other DSP components who have had less than stellar results with the on-board AVR 'room correction' products.

JSS 

[Kvalsvoll]

@SME, you are aware that loudspeaker manufacturers tune their speakers ("voicing") according to power response and estimated typical room acoustic properties, and have done so for decades?

If you look at on-axis measurements of typical hifi-speakers, they often have a response that deviates considerably from flat - look up some of the well regarded speakers on stereophile. This is because the radiation pattern  changes with frequency, and by adjusting the response the "sound" can be changed into more balanced and neutral, but the measured charts seemingly show a "defect" speaker.

Since most speakers have a pattern omni at low f and then narrowing towards higher f, they will radiate more power at lower frequencies. This causes some problems if you design a speaker with radically different pattern - a speaker with flat fr and flat power will sound different from these typical once, and since music is mastered for use with the typical speaker and room, it is necessary to alter the reposne of the flat speaker into something that gives a perceived tonal balance more similar to those typical ones.

[SME]

2 hours ago, maxmercy said:

So what would you do if you had what most people on forums have available: limited budgets, limited placement options, some room treatment, REW measurement capability and limited parametric/IIR correction capability?

Would you mainly focus on minimum phase problems in the LF and largely ignore MF/HF, or how would you go about optimizing a system with limitations like the above, given what you know now?

I ask not only because that's the way my system is set up, but many others with miniDSP or other DSP components who have had less than stellar results with the on-board AVR 'room correction' products.

Let's first assume that I have access to some future version of the tools I've created to do this sort of thing.  The tools would probably take the place of REW for measurement but could export WAV files that one could import into REW.  These tools use measurement data to obtain accurate estimates of first arrival and in-room power response, from which they use to suggest optimal correction curves.  From there, the problem would be to generate filters that could be implemented on more modest DSP hardware than I'm using now.

Some of the more recent MiniDSP products look pretty capable as far as offering FIRs along with a greater number of biquads than earlier models.  With mathematical optimization for the biquads, it may be possible to implement some pretty good corrections on the more capable MiniDSPs.  A complication is the fact that so many different MiniDSP hardware configurations are possible, each with somewhat different constraints.

Of course, my tools are not currently available for wider use, and I don't know when they will be or how affordable they'll be.  So what would I do without those tools?  That's a tough call.  An approach similar to Lygndorf's for estimating power response may work OK, but there are many details to be mindful of.  Randomizing the measurement locations in 3D space is probably critical, and I'd opt for more than 6 measurement locations if possible.  The different measurements should probably be level-matched before averaging.  Then there is the task of smoothing power response while keeping first arrival approximately flat, and it's important to prioritize fixing low Q features over high Q features of the same magnitude.  Contrary to common belief, low Q features are more audible than high Q features.  To get this right requires smoothing, and unfortunately, REW lacks power smoothing, which I believe is the best kind to use for this.  Most of the other stuff can probably be done in REW in some way or another.

What to do without a good power estimate?  At that point, falling back on existing advice is best.  Start with speakers that have an excellent native power (and on-axis) response.  Place them far from walls or install them flush-mounted, if possible.  If not, some absorption on the part(s) of the wall(s) closest to them may or may not help.  EQ can be experimented with but it's hard to know where and how much to apply.  Modal resonances may be a relatively easy problem to treat with EQ.  The peaks tend to stand out in the in-room measurements, and they can be particularly annoying to listen to.  Be sure to confirm they are actually modal resonances by looking at several measurements throughout the room.  Be willing to experiment to get the right EQ filter.  In my "optimized" system the modal resonances are only partially suppressed in most of the in-room measurements.  Unfortunately, none of this helps with lower Q stuff, including broad bass shape, that's critical to getting the best sound.  One can try to optimize broad shape by ear, which is what I did up until I applied my latest tools, which finally gave me a superior result than I could get by ear.

Sorry that I can't give a better answer than this right now.

[SME]

13 hours ago, Kvalsvoll said:

@SME, you are aware that loudspeaker manufacturers tune their speakers ("voicing") according to power response and estimated typical room acoustic properties, and have done so for decades?

If you look at on-axis measurements of typical hifi-speakers, they often have a response that deviates considerably from flat - look up some of the well regarded speakers on stereophile. This is because the radiation pattern  changes with frequency, and by adjusting the response the "sound" can be changed into more balanced and neutral, but the measured charts seemingly show a "defect" speaker.

Since most speakers have a pattern omni at low f and then narrowing towards higher f, they will radiate more power at lower frequencies. This causes some problems if you design a speaker with radically different pattern - a speaker with flat fr and flat power will sound different from these typical once, and since music is mastered for use with the typical speaker and room, it is necessary to alter the reposne of the flat speaker into something that gives a perceived tonal balance more similar to those typical ones.

Yep, I am absolutely aware of these things.  Almost no speaker sounds right with a ruler flat on-axis response.  Of course, that doesn't mean that manufacturers are tweaking their speakers on the basis of actual power response measurements, which are difficult to do correctly.  I'd bet that almost every speaker design gets tweaked at least once, based solely on subjective listening tests before being finalized.  My guess is that most speakers start up with fairly flat on-axis responses as a baseline and then get tweaked from there

There was a time when more speakers were designed specifically for flat power.  As I understand it, there was a kind of rivalry between people who believed the flat on-axis was optimal and those who believed that flat power was best.  JBL was a big proponent of the former.  Allison was a big proponent of the latter.  Allison did some remarkable work designing speakers for placement against walls and trying to optimize passive signal shaping to maintain nearly flat power under those conditions.  I grew up listening to one of Allison''s designs, which I remember fondly.

In the end, flat on-axis essentially won in listener preference experiments.  However, flat power may be have lost in part because such speakers almost always had up-sloping first arrival responses, for the reasons you note above.  Or maybe it's not about the up-sloping first arrival response but the *lack of slope in the power response*?  Personally, I'm pretty certain first arrival is still very important.   Otherwise, one would expect horns to sound rather dark compared to cone-and-dome speakers with the same on-axis FR, which is definitely not the case.

In any case, I have no doubt in my mind at this point that power response counts for a lot.

[Kvalsvoll]

Speaker tuning with radiation pattern, on-axis response, power response and then add in room acoustics - which will be more or less an unknown parameter for a speaker designer. This is difficult and complex, and has huge impact on perceived sound - in contrast to amplifiers, dac's, all the nonsense products.

I also believe that this field has not yet been fully discovered, there is still more to learn and find out.

Experiments focusing on how perception of sound relates to the technical parameters are key factors for improvement.

Horn speakers sound different from trad-hifi partly due to fundamental differences in radiation pattern. It is impossible to make them sound equal, because tonal balance depends on what sound is being reproduced. If you tune for flat and equal steady-state, the transients will sound different because decay profiles are different.