Keith_W DSP system

[Keith_W]

After a week of trying out different experiments with the driver phase linearization procedure, and generating a whole bunch of filters and listening:

- drivers individually amplitude and phase linearized with reverse AP filter
- drivers amplitude linearized only with no phase linearization with reverse AP filter
- SOME drivers amplitude linearized AND phase linearized, others left alone (given it is a 4-way system there are a lot of permutations! I settled on only phase linearizing the woofers and midrange horns)
- no driver linearization, only overall room correction with reverse AP filter

... I have concluded: there is a definite audible difference with phase linearization in any form, even if the frequency response looks the same. It is usually negative, it has the effect of dulling the sound, introducing a smearing quality, killing liveliness and dynamics. This is either due to the correction itself, or some artefact introduced by the correction which I have not recognized because I don't know where to look.

I have not found a "target curve" for phase, or even know what it should ideally look like. Perhaps one does not exist.

I am even more confused now, because phase is supposed to be inaudible. Yet I will swear under oath that I am not lying, the difference is easily audible. As mentioned, I don't know if this is the correction itself which is producing the audible difference, or an artefact introduced by the correction. I even tried doing a mono recording (both speakers playing together) to look for possible comb filtering which may account for the difference in tonality ... but nothing.

In the end, I made this decision: if you don't know what you are doing, it is better to do nothing. So I removed all phase corrections, and it sounds wonderful now. I mentioned upthread that without any phase linearization, it seems to "unleash" the speakers. They are so aggressive, and so dynamic. The "fatiguing" quality I noted earlier has gone after the filter "burned in" my ears. In fact, I find I prefer my correction over Dr. Uli's ... and the only difference between his correction and mine is that his has individual driver phase linearization.

 

[Music1969]

So I removed all phase corrections

so u mean no more individual driver time alignment ?

 

[GaryY]

After a week of trying out different experiments with the driver phase linearization procedure, and generating a whole bunch of filters and listening:

- drivers individually amplitude and phase linearized with reverse AP filter
- drivers amplitude linearized only with no phase linearization with reverse AP filter
- SOME drivers amplitude linearized AND phase linearized, others left alone (given it is a 4-way system there are a lot of permutations! I settled on only phase linearizing the woofers and midrange horns)
- no driver linearization, only overall room correction with reverse AP filter

... I have concluded: there is a definite audible difference with phase linearization in any form, even if the frequency response looks the same. It is usually negative, it has the effect of dulling the sound, introducing a smearing quality, killing liveliness and dynamics. This is either due to the correction itself, or some artefact introduced by the correction which I have not recognized because I don't know where to look.

I have not found a "target curve" for phase, or even know what it should ideally look like. Perhaps one does not exist.

I am even more confused now, because phase is supposed to be inaudible. Yet I will swear under oath that I am not lying, the difference is easily audible. As mentioned, I don't know if this is the correction itself which is producing the audible difference, or an artefact introduced by the correction. I even tried doing a mono recording (both speakers playing together) to look for possible comb filtering which may account for the difference in tonality ... but nothing.

In the end, I made this decision: if you don't know what you are doing, it is better to do nothing. So I removed all phase corrections, and it sounds wonderful now. I mentioned upthread that without any phase linearization, it seems to "unleash" the speakers. They are so aggressive, and so dynamic. The "fatiguing" quality I noted earlier has gone after the filter "burned in" my ears. In fact, I find I prefer my correction over Dr. Uli's ... and the only difference between his correction and mine is that his has individual driver phase linearization.

I don't have good enough knowledge at all, but sometimes I think what is the sense of phase alighment at 1.7kHz or above because the distance of 2 ears is more than wavelenghth....

 

[3ll3d00d]

I have not found a "target curve" for phase, or even know what it should ideally look like. Perhaps one does not exist.

It was commented on earlier in the thread, it's typically just a minimum phase response which is what acourate aims for itself and which helps you get that nice sharp step response (which is also what people typically look at when reviewing acourate correction results).

 

[OCA]

Here's a properly produced minimum phase version of Dr Toole target curve for you.
If you're too desperate, you should invert speaker responses over this one.

 

[ernestcarl]

... I have concluded: there is a definite audible difference with phase linearization in any form, even if the frequency response looks the same. It is usually negative, it has the effect of dulling the sound, introducing a smearing quality, killing liveliness and dynamics. This is either due to the correction itself, or some artefact introduced by the correction which I have not recognized because I don't know where to look.

There were signs of "over correction" in the step response: teeny-tiny (likely very subtle) pre-ringing perturbations.

-----------

I've recently converted my bass-limited coaxial center speaker from a 2-way single-amp design to 2-way bi-amp... but kept the passive xo to protect the HF horn driver. Why? Boredome and curiousity got the better of me...

Here's some recent results (seems a little bit better from prior, I think) of the sub+center coax xo sum:



The one which produces the most audible "artifacting" from phase linearization should be obvious: it's the last wavelet plot! But this is mainly because of the limited LF extension of this ported speaker where it cannot fully cancel out the pre-echo with the sub itself. Do note, however, that I cannot hear any pre-echo or obvious smearing in the ~40ms FIR version (second worst looking plot) despite the visible pre-causal time-smear -- Although, 60 dB scale in a spectrogram is really nitpicking with a microscope. Yeah, I did my usual bass kick-drum testing and practical real-world pop music listening... usual tracks with a clear bass line...

Last two plots has ~40ms FIR (processing latency accounting for all channels in 7.1 nearfield desk MCH system) -- in reality, it's only 27ms for the sub and 37ms for the coax itself.

~0.7m distance to coaxial speaker

The phase EQ filtering in this case is actually pretty "simple":



But, actually, the min phase filter design is far more important:

 

[Keith_W]

Here's a properly produced minimum phase version of Dr Toole target curve for you.
If you're too desperate, you should invert speaker responses over this one.

Thanks. I imported it to take a look. In red is the target curve you posted, in green is the one I am currently using, courtesy of Dr. Uli. It sounds superb. I have tried a similar looking target curve to yours before:

Yours in red, my old target curve in brown.

The difference between Dr. Uli's curve and my old brown curve is quite different tonality, as you would expect. Dr. Uli's curve is much fatter in the midrange which really gives instruments a lot of nice body and weight. My old brown curve sounded flat and detailed with a lot of bass oomph. It is surprising what a couple of dB can make.

I will give yours a try and let you know what I think

 

[Keith_W]

so u mean no more individual driver time alignment ?

There is individual time alignment, just no phase linearization. I would never remove the time alignment, it is crucial in my system. Without time alignment, the bass sounds disconnected and flabby. The entire reason I embarked on this DSP journey was to fix the bass. At first, I thought it was a matter of amplitude - add a subwoofer, and i'm done. But years of mistakes and mis-spent money has lead me to this point. Of course I am still making mistakes, and I post it here for all to see. Hopefully we all learn from it.

 

[Keith_W]

View attachment 357473 View attachment 357474 View attachment 357476

There were signs of "over correction" in the step response: teeny-tiny (likely very subtle) pre-ringing perturbations.

Yep, I saw that in a zoomed in view of the step response as well. This is a zoomed in view of the step response of the filter with individual driver phase linearization, left and right channel:

And this is a zoomed in view of the step response of the filters with no phase linearization:

It doesn't look too different to me, yet one sounds noticeably clearer than the other.

Someone mentioned upthread that there may be post-ringing which causes the smearing effect. I tried looking for that as well, including examining the filters (instead of the measurements).

This is a comparison of the woofer crossover of the left channel only. In red, we have individual driver phase linearization. In green, no phase linearization. That's right, the horrible looking green filter is actually the clearer sounding filter.

At this point, I gave up. I don't understand why the uglier looking filter should sound better. Or that my corrections, which on paper look nicer than Dr. Uli's, don't sound as good as his. And I can imagine that it would be frustrating for all you guys on ASR as well, because you have to take my word for what it sounds like instead of being here to listen to it for yourself.

Here's some recent results (seems a little bit better from prior, I think) of the sub+center coax xo sum:

View attachment 357485 View attachment 357486View attachment 357487View attachment 357488 View attachment 357489

The one which produces the most audible "artifacting" from phase linearization should be obvious: it's the last wavelet plot! But this is mainly because of the limited LF extension of this ported speaker where it cannot fully cancel out the pre-echo with the sub itself. Do note, however, that I cannot hear any pre-echo or obvious smearing in the ~40ms FIR version (second worst looking plot) despite the visible pre-causal time-smear -- Although, 60 dB scale in a spectrogram is really nitpicking with a microscope. Yeah, I did my usual bass kick-drum testing and practical real-world pop music listening... usual tracks with a clear bass line...

I guess the difference between your filters and mine is that your artefacts are not audible. Mine are. And as I have shown above, both filters have similar looking artefacts, so I am not confident that a small degree of pre-ringing is the cause of all the smearing.

The above video demonstrates pre-ringing with my speakers. I can hear it on my laptop speakers so it should be easily audible on whatever device you are using. This is not a current filter, it was one that I made several months ago. Listen for the short crescendo-like hum which occurs before each drum strike. It goes "mmm-DUMMM" instead of "DUMMM".

So I use that particular track to listen for pre-ringing - Wadaiko Matsuriza Japanese Drums. It's available on Tidal and maybe some other services.

For smearing I use orchestral work like the third movement of Mahler's 9th Symphony - Simon Rattle. This is an especially complex and chaotic orchestral work. When all instruments are playing, it should be possible to isolate individual instruments and clearly hear the timbre. With my "good" filter, I can do this. With my bad filter, it sounds all smeared together, a bit like my clock radio.

 

[OCA]

View attachment 357608

Thanks. I imported it to take a look. In red is the target curve you posted, in green is the one I am currently using, courtesy of Dr. Uli. It sounds superb. I have tried a similar looking target curve to yours before:

View attachment 357611

Yours in red, my old target curve in brown.

The difference between Dr. Uli's curve and my old brown curve is quite different tonality, as you would expect. Dr. Uli's curve is much fatter in the midrange which really gives instruments a lot of nice body and weight. My old brown curve sounded flat and detailed with a lot of bass oomph. It is surprising what a couple of dB can make.

I will give yours a try and let you know what I think

I don't touch above 200Hz much but I am quite impressed with the even less bass heavy TC of Dr Uli's. They say trained ears need less bass. I am getting a beating online at the moment because I forced Dr Toole's curve in the Audyssey automation script and apparently it's not as as "bombastic" as the Harmon

 

[dualazmak]

Of course, I still keep watching and learning a lot on this invaluable thread; thank you again Keith for your sharing intensive measurements on your DSP efforts!

BTW, let me just share my general feeling/thoughts after reviewing the recent posts and discussions kindly shared during the past two weeks or so.

I assume, at least for my eyes, that your are sticking to do all of (1) time alignment, (2) phase tuning (almost abandoned?), (3) possible EQs, and (4) tonality tuning, in upstream digital domain by single DSP software.

Even though I fully understand your intensive efforts in this direction, I would like to suggest that you may test and evaluate optimal compromised combination of (simpler or simplest) DSP tunings (including time alignment, of course) and analog-level tunings, especially flexible on-the-fly tonality controls in analog-level.

For your preference reference, after my rather intensive efforts in DSP multichannel audio system, now I believe that flexible relative-gain controls in analog-level by using (a) analog-output gain control of multichannel DAC, (b) possible use of gain controller of active subwoofer(s), (c) possible use of multiple pre-amplifier, and/or (d) possible use of multiple integrated amplifier, would give you much more "freedom" for your safe and flexible on-the-fly tonality tuning depending on e.g. the genre of the music, recording quality of the music track, hearing preferences of audience(s), and/or hearing capability of the audience(s) especially the possible age dependent hearing decline in high frequency. As you can well understand, these relative gain (tonality) controls in analog-level has no effect on upstream DSP configurations.

At least in my case, I prefer safe and flexible on-the-fly relative gain (tonality) controls by using multiple integrated amplifier while keeping the relative gain (relative sound volume) of L&R subwoofers at a constant level; the master volume is, should be, controlled by most upstream digital music player (in my case JRiver MC).

Spoiler

Please refer e.g. here here #438 and here #643 for example cases of safe and flexible on-the-fly analog-level relative gain (tonality) controls.

The standard/default gain values in the diagram are for my standard/routine classical music listening.
I often gain up the midrange by changing the gain of ACCUPHASE E-460 from -17.0 dB to -13.6 dB, i.e. 3.4 dB boost, when I enjoy typical smooth jazz trio tracks, like albums of Karel Boehlee Trio (ref. here #640)...

 

[3ll3d00d]

At least in my case, I prefer safe and flexible on-the-fly relative gain (tonality) controls by using multiple integrated amplifier while keeping the relative gain (relative sound volume) of L&R subwoofers at a constant level; the master volume is, should be, controlled by most upstream digital music player (in my case JRiver MC).

Given you can do all of this precisely in software, this looks like a preference for physical knobs and dials. If it's not that, what practical advantage do you think this would provide?

 

[dualazmak]

Given you can do all of this precisely in software, this looks like a preference for physical knobs and dials. If it's not that, what practical advantage do you think this would provide?

I know that XOs (many parameters), time-alignment, phase tuning, EQ and gain in DSP would be more-or-less interdependent with each other, right?
None of them can be changed/modified completely independently.
On the other hand, analog-level gain controls do not affect the upstream DSP configurations.

One of the other pros of HiFi analog-level relative gain tuning (tonality control) by knobs/dials would be it is very safe on-the-fly compared to on-the-fly DSP gain control.

In case if you would like to do it in DSP, especially on-the-fly, you always have possibility of mis-adjustment of gains, i.e. and e.g. 20 dB boost instead of intended 2.0 dB boost by numerical keyboard mistyping the value (and/or it would happen even using mouse wheel up-and-down in some DSP software tools) which may harm and/or destroy your precious SP drivers.

Consequently, I seldom change DSP parameters (especially relative gains) on-the-fly, while listening to music track; I always stop playing and set the master volume in digital music player to minimum (minus infinity dB) position when changing the DSP parameter(s), and after the DSP modification and after start playing music, I very carefully and slowly gain-up the master volume (in my case JRiver MC) to check the given change in DSP parameters; in this way, I can avoid possible (but rare) harm/damage to SP drivers.

For further safety purposes, I also use protection capacitors for my treasure Be-midranges, Be-tweeters and metal-horn super-tweeters (summary ref. here).

I use DSP EKIO, and EKIO has very nice gain up-and-down by mouse wheel rotation (0.1 dB granularity); I have been always strongly recommending, therefore, to use mouse wheel rotation (not keyboard numeric typing), if you use EKIO and would like to do on-the-fly relative gain control (e.g. ref. here).
(Mr. Guillaume BADAUT of LUPISOFT who develops EKIO has very quickly implemented the excellent mouse wheel operations responding to my sincere request.)

 

[ernestcarl]

Last two plots has ~40ms FIR (processing latency accounting for all channels in 7.1 nearfield desk MCH system) -- in reality, it's only 27ms for the sub and 37ms for the coax itself.

~0.7m distance to coaxial speaker

I noted this PA speaker -- purchased on ebay for cheap -- was "out-of-spec". Well, I consulted the B&C 6FHX51 spec sheet today and it turns out I was wrong:

*Ignore the unevenness below 1kHz as that's largely because of room/desk "boundary effects"

There is a dip there after all! I haven't looked at those graphs too carefully and it's been months. I previously thought the HF horn driver filled-in for the LF driver, but apparently not so completely when the passive xo is in place. Although, there is still the possibility that there is still something wrong with the xo board or perhaps even HF horn driver itself.

6FHX51 8 Ω - Coaxials - B&C Speakers

www.bcspeakers.com

The difference in sensitivity in the drivers is very stark, to say the least.

Given that bi-amping gives one independent control for each driver section, it does seem to make good sense to EQ with that in mind:


*HF horn section still has passive filter network in place

Levels are controlled digitally in more than one place: first inside JRiver and and second in a miniDSP 2x4HD.

Unlike @dualazmak, I prefer to keep the amplifier with its controls well out of sight inside a utility room. Mind you, it's also by necessity as the fan is so noisy -- again, second-hand and bought for cheap! I'd rather not allow changes to the gain settings on the device and cover the thing to try "fix it" permanently in place. On and off switching is controlled via a remote switch right beside the speaker itself.



It's not really possible to over-drive the speaker as I've set the gain staging carefully so that even at 0 dBFS the speaker can handle it:



In JRiver if I put in a volume level value above 0 dBFS deliberately in the PEQ section, it actually does not play any louder -- besides the soft clipping function, it is just not possible to play louder than 0 dBFS. I've tried putting in +5 dBFS and +10 dBFS but the measured SPL level will not rise anymore.

You could still clip your amp and/or external DAC/processor elsewhere in the chain, of course. It's just that I've deliberately limited the amount of gain boost in the amps so that even if I accidentally max everything out by accident in software, nothing should blow out -- assuming one does not leave max SPL playback indefinitely.

 

[dualazmak]

Unlike @dualazmak, I prefer to keep the amplifier with its controls well out of sight inside a utility room. Mind you, it's also by necessity as the fan is so noisy -- again, second-hand and bought for cheap! I'd rather not allow changes to the gain settings on the device and cover the thing to try "fix it" permanently in place. On and off switching is controlled via a remote switch right beside the speaker itself.

Thank you for sharing your policy and approach which I can well understand especially if using amplifier(s) with fan(s) of unacceptable fan-noises.

Spoiler

Fortunately, in my system/gears, all of these, including the two audio dedicated PCs (ref. here), are completely silent:

Please refer here for the details of my latest system setup.

I also understand and agree on the feasibility that you have implemented several "safety gain limitters" in digital music player and/or DSP.

Yes, we can have several feasible/applicable policies and/or approaches in DSP-based multichannel audio system depending on personal preferences, audio gears, DSP (and analog) tuning preferences, music genre, prefered (or not) on-the-fly relative gain (tonality) tuning, room acoustics, and so on.

I believe, for many people aiming towards DSP multichannel system, it would be very nice and useful that we can share various DSP policies and approaches (like yours, Keith's and mine) on this thread and on other threads in ASR Forum!

At least in my case, I still prefer compromised optimal combination of simple (or simplest) DSP configuration and analog-level safe and flexible relative gain control/tuning as described in above post #311.

 

[Keith_W]

At least in my case, I still prefer compromised optimal combination of simple (or simplest) DSP configuration and analog-level safe and flexible relative gain control/tuning as described in above post #311.

I have that too. I don't think you could build yourself an active DSP system without gain control on each channel, particularly in my case. My horns, as delivered, have a resistor in the XO circuit to drop the sensitivity down to the level of the woofer. I have modified my speakers by (1) bypassing the horn crossover, which increases the sensitivity to 110dB/W/m, and (2) swapping the woofer for a custom prototype, which decreased the sensitivity to 85dB/W/m.

IOW there is a 25dB difference between the horn and the woofer, which would require a massive cut in the filters. On top of that, I use digital volume control, which means I usually listen another 30-40dB below that. The noise floor of my DAC suddenly goes from -110dB to -45dB. Now this is likely not audible or at least very difficult to hear, but I prefer to push the noise floor lower. By having analog trim, I can cut back the horn by the required 30dB and thus push the DAC's noise floor down by another 25dB so that it is really inaudible.

Of course it is possible to tweak the volume of each driver on the fly by simply twiddling the knobs on each volume trim, but why would I want to do that? I carefully set up the system by measuring the volume of each driver and adjusted the trim. Then I taped over the volume controls and I don't touch them under any circumstances.

 

[dualazmak]

Thank you for your invaluable response and comments to me which should be also nice reference and of interest for many people following this wonderful thread!

Of course it is possible to tweak the volume of each driver on the fly by simply twiddling the knobs on each volume trim, but why would I want to do that? I carefully set up the system by measuring the volume of each driver and adjusted the trim. Then I taped over the volume controls and I don't touch them under any circumstances.

Again, I well understand and highly respect your above policy and approach of "fix and tape" the relative gain knobs in analog amplifier level.

As I repeatedly shared, however, at least in my case, I still would like to have my more-or-less freedom in relative gain (tonality) control on-the-fly at analog level, even though I usually "fix" (but I do not tape) each of the gain knobs of my four integrated amplifiers at my standard/default positions optimized for my daily classical music listening. Of course, I too have some relative gain adjustments also in my DSP configuration.

Spoiler

I occasionally further fine tune, however, the relative gains (tonality) (within about +/- 5 dB range) in analog amplifier level depending on genre of music, recording quality, age-dependent hearing decline of specific audience(s), etc.

Just for example, when I enjoy my beloved smooth jazz trio music albums of Karel Boehlee Trio (rather soft-side recordings, ref. here), I boost-up my midrange (driven by ACCUPHASE E-460) about 3.4 dB for best fit to my jazz listening preference. This is just my personal preference and "freedom", even though I well understand that some (or many?) people here in ASR Forum would criticize (or blame) me by saying "such a tonality tuning depending on genre (or on recording quality) of music tracks would be blasphemy or heresy in HiFi audio listening!" , but I do not care it, please let me enjoy music based on my preferences and my tuning style!

Spoiler

You would please find a typical case here:
- A serious jazz fanatic friend came to my home for audio sessions using my multichannel multi-driver multi-way multi-amplifier stereo system: #438

The flexible on-the-fly tonality fine tuning (especially in high-Fq zone covered by tweeters and super-tweeters) compensating possible age-dependent slight hearing decline would be a little more serious issue when I would invite "various" music-lover guests to our audio listening session at my listening room.
Even for myself and my wife, since we know the diagnostically-proven slight hearing decline above 7 kHz, I (we) prefer a slightly upward SPL for 7 kHz to 20 kHz.
Again, this is just my personal preference and "freedom", even though I well understand that some (or many?) people here in ASR Forum would criticize (or blame) me by saying "such a tonality tuning depending on hearing capabilities would be blasphemy or heresy in HiFi audio listening!" , but I do not care it; please let me enjoy music based on my preferences and my tuning style!

Spoiler

- Excellent Recording Quality Music Albums/Tracks for Subjective (and Possibly Objective) Test/Check/Tuning of Multichannel Multi-Driver Multi-Way Multi-Amplifier Time-Aligned Active Stereo Audio System and Room Acoustics; at least a Portion and/or One Track being Analyzed by Color Spectrum of Adobe Audition in Common Parameters: [Part-11] Violin Music: #643

And, you would please let me describe this paragraph again here:
I know that XOs (many parameters), time-alignment, phase tuning, EQ and gain in DSP would be more-or-less interdependent with each other, right? None of them can be changed/modified completely independently. On the other hand, analog-level gain controls do not affect the upstream DSP configurations.

 

[gnarly]

After a week of trying out different experiments with the driver phase linearization procedure, and generating a whole bunch of filters and listening:

- drivers individually amplitude and phase linearized with reverse AP filter
- drivers amplitude linearized only with no phase linearization with reverse AP filter
- SOME drivers amplitude linearized AND phase linearized, others left alone (given it is a 4-way system there are a lot of permutations! I settled on only phase linearizing the woofers and midrange horns)
- no driver linearization, only overall room correction with reverse AP filter

... I have concluded: there is a definite audible difference with phase linearization in any form, even if the frequency response looks the same. It is usually negative, it has the effect of dulling the sound, introducing a smearing quality, killing liveliness and dynamics. This is either due to the correction itself, or some artefact introduced by the correction which I have not recognized because I don't know where to look.

I have not found a "target curve" for phase, or even know what it should ideally look like. Perhaps one does not exist.

I am even more confused now, because phase is supposed to be inaudible. Yet I will swear under oath that I am not lying, the difference is easily audible. As mentioned, I don't know if this is the correction itself which is producing the audible difference, or an artefact introduced by the correction. I even tried doing a mono recording (both speakers playing together) to look for possible comb filtering which may account for the difference in tonality ... but nothing.

In the end, I made this decision: if you don't know what you are doing, it is better to do nothing. So I removed all phase corrections, and it sounds wonderful now. I mentioned upthread that without any phase linearization, it seems to "unleash" the speakers. They are so aggressive, and so dynamic. The "fatiguing" quality I noted earlier has gone after the filter "burned in" my ears. In fact, I find I prefer my correction over Dr. Uli's ... and the only difference between his correction and mine is that his has individual driver phase linearization.

I so agree with you about "if you don't know what you are doing, it is better to do nothing."

That's the way i feel about room corrections,..... they still feel over my head despite me feeling extremely comfortable tuning/processing DIY speakers with multi-channel linear-phase FIR DSP...in a quasi-anechoic environment.

Once the room enters the equation, I want to separate whatever I've achieved with speaker tuning, from what I would do for room corrections.
Seems the speaker should be what it is, no matter what room I put it in.

I think trying to make lump-sum corrections of both the speaker and the room is the most likely root cause of room corrections not working as well as hoped.
Simply because such corrections invariably amplify focus on a single location, or at best a narrow location area.


I think whatever went into speaker processing to achieve the smoothest and flattest mag on phase over as wide an area as possible, should be left alone.
Iow, I think it's best not to do any in-room driver by driver type work. Or linearize IIR xover rotations with inverse all pass.
(If the xover rotation being linearized is the acoustic measurement, it becomes one of a stronger focus to a spot, than using linear-phase crossovers to begin with, ime.)

So I really don't like the idea of combining both speaker tuning and room correction, into one set of FIR files handling both duties.
The few times I've tried that with FIR global correction, I've had similar experiences/frustrations to some that you mentioned.
Whereas I almost never get disappointed with full out flat mag and phase via FIR, applied to just the speaker (as best as possible)

just my 2c

 

[OCA]

Major update to REW Beta today. Action windows gone, multiple measurements can be moved around with mouse, most menus show up with right click, countless improvements here and there but will surely require some adaptation time especially for seasoned REW experts like @ernestcarl :

REW Beta Release - REW API beta releases

Feature request: Please add support for M-noise ! https://m-noise.org/ It's a really great way to test modern systems that have a DSP compressor, putting devices that don't have a DSP on a level playing field with those who do. It enables to fully asses the linear range of a speaker system when...

www.avnirvana.com

 

[ernestcarl]

Major update to REW Beta today. Action windows gone, multiple measurements can be moved around with mouse, most menus show up with right click, countless improvements here and there but will surely require some adaptation time especially for seasoned REW experts like @ernestcarl :

REW Beta Release - REW API beta releases

Feature request: Please add support for M-noise ! https://m-noise.org/ It's a really great way to test modern systems that have a DSP compressor, putting devices that don't have a DSP on a level playing field with those who do. It enables to fully asses the linear range of a speaker system when...

www.avnirvana.com

Funny -- seasoned (maybe), not exactly "expert" as I don't even have a good grasp of some of its other features. I've actually been using the more "stable" standard release version recently in my old Win 10 laptop which I prefer to use for quick REW measurements over Linux. Now, I am a little piqued by the beta changes so will take a peak. To be honest, it took me quite a few months to finally get used to some of the UI changes like the new PEQ window interface so I've kept two software versions installed... it's always jarring when you can't find buttons in their "original" locations.