Testing room correction systems

[markus]

Ever so often the question comes up which room correction system (RCS) is "best". In order to explore this topic this thread was created.

First let's see if there's common ground on which we can move forward.


In-room measurements vs. perception vs. equalization

At low frequencies the (acoustically small) room dominates what we hear so the premise is that a steady state measurement lumping together direct and reflected sound highly correlates with what is heard up to the sub/sat crossover region.
Equalization here can improve the frequency response. It can also improve ringing. If used incorrectly it can make ringing worse though.

At higher frequencies reflections have a significant impact on what is measured but not on what is heard (we hear added spaciousness, not coloration). Thus any attempt at equalizing the steady state response would alter the direct signal (and reflections) spectrally and likely would make it sound worse.
Equalization could still be beneficial if reflections could be "removed" from measurements taken at the listening position. The result would be the anechoic response of the speaker hence room correction would become speaker correction. Depending on the quality of the speaker this could improve the response quite a bit.
Reflections can be "removed" from measurements to a certain degree by gating techniques. The problem is that there are very early reflections from objects close to the speaker and the microphone that have a significant impact on the measured results despite gating.


Measuring RCS performance

Based on what has been said above there's performance criteria in two areas:
1. Low frequency performance ("room correction")
1.1. Magnitude response - how good does the equalized result follow the target curve?
1.2. Ringing - has ringing been reduced?
2. High frequency performance ("speaker correction")
2.1. Magnitude response - is the anechoic direct sound from the speaker closer to flat/target curve?


Testing

General
1. Room correction needs to improve ALL points within the listening area so multiple measurements need to be taken before and after a specific room correction system is run. These control points should NOT be the same as the mic positions used by the room correction system. This allows us to assess whever the room correction system improved the whole listening area or just the small number of single points used by the room correction system while other points have been made worse.
2. Mics positioned at control points can't be moved between testing otherwise no comparative assessment between room correction systems can be made. Even small mic location variations will create significantly different results.
3. Anechoic measurements of the speaker need to be performed with and without room correction enabled (for evaluation of higher frequencies).

Evaluation
1. Low frequencies
1.1. How close does each control point follow the target curve? (Magnitude response)
1.2. Has the spread between control points been reduced/made worse? (Magnitude response)
1.3. Has ringing been reduced/made worse? (Waterfalls)
2. Higher frequencies
2.1. Has the direct sound been improved/made worse? (Magnitude response)
2.2. Has the phase relationship between stereo pairs (L/R, Lss/Rss, Lsr/Rsr, etc.) been improved/made worse? (Phase response)

 

[RichB]

This is an interesting topic.

Thus far, ASR measurements have targeted the base-level performance of processors measuring them and comparing them to simpler 2-channel test.
I think the next logical step is to understand the performance of the DUT when the DSP signal path is engaged.

For example, all curves set to flat with the curtain or upper limit set to say 50Hz and provide the same measurements as the Pure Direct path.
This provides a baseline performance with processing engaged.
It could be repeated with PEQ as well which may be a different path.

- Rich

 

[Zooqu1ko]

2. Mics positioned at control points can't be moved between testing otherwise no comparative assessment between room correction systems can be made. Even small mic location variations will create significantly different results.
3. Anechoic measurements of the speaker need to be performed with and without room correction enabled (for evaluation of higher frequencies).

Both of these could be worked around if, instead of a real microphone, the room response was only simulated. First, the test signal would have to be routed to a computer, altered according to a real measured room impulse response and fed back into the room correction system in near real time. Alternatively, a simulated room could be used, which would be especially useful to systematically tease out information how the room correction system reacts to specific changes in the room characteristics.

After the room correction system has been set up, the same room simulation could be used to reliably characterize the "improvement" the RCS provides.

While all that may be manageable with systems that can work with arbitrary calibrated microphones, those like Trinnov or even Audyssey that require "special" microphones would make it much more difficult to splice the simulated room response intauo the appropriate inputs.

 

[markus]

Another idea: convolve the transfer function of the preamp output after correction with the transfer function of the control points to obtain the corrected in-room response. This way the control points need to be measured only one time. So no issue with variability of mic placement. Furthermore the convolution can be redone whenever a new RCS (version) is measured.

 

[Pio2001]

I Don't think that using simulations is a good idea.
Assuming that the simulation works means assuming that the room correction system works perfectly.

Since we want to test if the room correction works, we can't assume this.

For waterfall, simulations might not work very well because of background noise during the measurement.
Even for amplitude response, I've got some unexpected differences between the predicted result and the actual result when I create a correction with REW.

 

[Soniclife]

I think this is an interesting topic, you regularly see people disliking RC after applying it, thin sound is the usual criticism, but they don't test blind, or level matched. The level matching is tricky, how do you match levels when the response can differ a lot on purpose? Pick a reference mid range frequency that hasn't changed, or use something like C weighting and pink noise?

 

[Jimbob]

I typically level match overall/or if different curves the midrange based on REW graphs with full smoothing (1/1), but I appreciate that this gets trickier as the freq response curves get more extreme.

 

[NoMoFoNo]

I'm afraid to ask how competent the room correction schemes in AVRs can be. I've owned several over the years that included the mic and spit out all sorts of tones, pink noise, white noise, etc, to correct for the setting. My latest, a Pioneer AVR, seems to do a nice job to my ears, but I'd love to learn more about correction.

 

[digicidal]

Thought I'd throw up some visual aides for the discussion as well as a very unscientific analysis of 3 approaches in my office. I say "unscientific" because of the vast number of uncontrolled variables - I did try to introduce as few additional ones as possible... but I've already taken a couple hours out of my work schedule to do this much. So take it with the large salt grain it deserves.

Office details:
A monumental pile of shit ahem, amount of work materials, spare parts, and embarrassingly large amount of "to do projects" in various stages.
Size is approximately 15' x 14' x 11' (4.5m x 4.25m x 3.35m or roughly 64m³ for the rest of the world)
Source: PC via Toslink MB output from ROG Rampage VI Extreme (Realtek S1220)
PrePro: Marantz 8801
Amps: Emotiva PA-1 (IcePower ASC300)
Speakers: Polk LSiM703 Bookshelf (decoupled a bit with half of an ISO-130 each)
Cables: who cares? Some crap I dug up. 14ga on speakers (runs are ~6ft ea) and XLR probably from monoprice or guitar center's cheapest... I forget.

To give you an idea just how bad my office is currently... some shameful photos:
<-- Audio(ish) side and working/listening side -->
Panels are some DIY 2'x2' and purchased 2'x3x on the others (Corning 703) - four of each.
So yeah... a "pristine listening environment" it's not. Oh and my main workstation has 5 radiators and ~40 fans on it as well as two pumps and a bit of coil whine which wasn't present due to my GPUs doing nothing during this time. Noise floor as measured at listening position was between 48.9dB and 53.6dB

As I sit behind the three monitors, I've aimed the speakers directly at my chair... which helps compensate some HF blocking they cause. All measurements for each of the three RC methods were performed "identically" (by putting a large piece of graph paper on my central monitor for Z coordinates and one below on the desk for XY). It would be a literal miracle if any of them were truly identical though... so grain of salt again.
Measurements were taken in the following order for each method - no additional points taken in any case (despite capability):
1) Ear height, normal head position.
2) 6" to the right of #1
3) 6" to the left of #1
4) 6" above #1
5) 6" below #1
6) 6" behind #1
7) Repeat of position #1 exactly.

The room correction methods were performed in the following manner:
Audyssey MultEQXT32 (&Flat) - mic positions were the same as above, but with the crappy D/M mic held in a clamp facing up.
REW & Dirac used identical positions - not sure how problematic the "preset positions" in Dirac were... I picked the closest in the Chair version to the positions listed above. Didn't seem to skew it too much so I don't think the physical position is weighted much... which makes sense as it's unlikely to be accurate to an image on a screen anyway.

For the REW measurements I took with Equalizer APO 1.2.1 handling everything except Audyssey (obviously). Preamp for peak matching and then two includes... one which has the Dirac VST and one with the REW auto-optimized for close to a Harman target curve - these were with the 8801 set to "Pure Direct" mode. Then turned off both and manually set the 8801 to Audyssey and Audyssey Flat - Dynamic EQ enabled but Dynamic Volume Disabled. Uncorrected was taken in Pure Direct but with Equalizer APO turned off for all elements.

Pictures for all who stuck around (graphs zoomed showing just 50dB range to exaggerate differences - but this matches my noise floor anyway) :

REW Optimized to house curve:

DiracLive VST:
- REMOVED - See bottom notes.
Audyssey MultEQXT32:

Audyssey MultEQXT32 (Flat):

Because I shrunk a bit... Red=REW Optimized Green=DiracLive VST Purple=Audyssey Orange=Audyssey Flat Black=Uncorrected

Random screens...
Dirac Live analysis and curve:

And doing some "critical listening" with Equalizer APO up for switching:

Subjective nonsense:
Honestly, they all sound much better than uncorrected - which (also honestly) sounds pretty damn good!
Listened to the following tracks:
Tool -"Pneuma" from Fear Inoculum
Steely Dan - "Night By Night" from Pretzel Logic
Valley of the Sun - "Gaia Creates" from Old Gods
Alan Parsons Project - "I Robot" from I Robot
Charles Fambrough - "Our Father Who Art Blakey" from The Proper Angle

Order of preference: REW Optimized, Dirac, Audyssey (std), Audyssey (flat), No DSP - but although the differences were apparent... only the last two were noticeably bad in a few areas. The REW optimization, Dirac and Audyssey Standard were all great sounding without much reservation. With Audyssey in the "flat" setting there was quite a bit of unrealistic "sizzle" on the cymbals - especially with Charles Fambrough. Even with the monitors throwing the block for me... my ears weren't happy. The rest sounded good however, despite Audyssey clearly not being as measurably good in some areas. Probably if I didn't have the monitors... highs would be too much - but this is the first implementation of MultEQXT32... I'd expect better on 8805 - and if you just had it handle the bass integration, it would be more than "liveable" IMO.

Without any DSP enabled of any kind the room modes were overwhelming in the 100-300Hz area (likely made worse with bounce off the credenza top as well). Although it gave some drums a bit more impact, it made everything else very muddy and congested sounding in the bass region. Above that everything sounded just a bit less "right" but honestly not hard to listen to at all... the speakers really are quite good! If they were on stands and apart from the walls more, and coupled with a sub... they'd be incredible and basically only need bass EQ at all.

Thanks for sticking with it... if you did not what you could even remotely call decisive or demonstrative even - my room, my issues after all. YMMV. And if you hadn't already... check what Audyssey achieved in the sub bass... that's on speakers that shouldn't go that low normally. Really a mixed bag... magic bass, but also obviously targeting a flat target even on the non-flat setting IMO.

EDIT: There's a huge problem with Dirac (in my implementation at least) VST utilization via Equalizer APO. It's on multiple levels and makes it basically useless in this case. Note that I don't mean useless overall... just useless in broad application. Unlike having an inline DSP unit to send the correction to, each spawned instance of the control is dynamically created as a unique device. The problem with this is twofold: since Equalizer APO doesn't have it's own playback capabilities (or I've missed them entirely) there's no way to perform calibration with it's control - I get the "cannot sync with device" error in those cases. If I used the Windows Audio subsystem... I'm calibrating a different control... which I can push the filter to, but then not necessarily use (because it spawns with a different device ID). A similar problem exists with REW (as far as Dirac is concerned) - I can't push the filter to it's instance, because it's somehow different (Dirac Live sees two copies of the same device ID in this case).

Finally I attempted to push a correction set to the Windows system (via playing a wav via the alerts sound configuration menu) which worked - I was able to sync, take all of my measurements, generate my filter... and bluescreen my system upon pushing the filter. That's it - I'm done for now. So I've pulled out the Dirac graph, it's meaningless in this context (thumbnail is below if you're interested).

I did leave my subjective impressions, because it's clear that the control in Foobar works and has my filter in it. The sound changes significantly when I switch to it, I can see my filter listed in it's VST console, etc. I just can't have that same filter affect REW for whatever reason, nor can I push the correction to a different Dirac device (which every instance is)! When I finish up a few projects that are partially responsible for the state of my office... I'll have more time to dedicate to figuring out where I've gone wrong, and likely do a fresh install on a dedicated system or laptop for further testing. Better yet, maybe just get a miniDSP box so there's a physical device rather than a virtual one - I'm thinking the VST solution is not fully mature at this point.

 

[digicidal]

I guess it's safe to say that they all do something well, and other things not as well. Dirac is really the best however in not overdoing anything, while the REW auto optimization to target is really good at doing a little of everything (and the only one that's not totally a "black box"). Audyssey's biggest flaw (in this implementation at least) is that it tries too hard to get to that curve... in some cases boosting by a large amount. Basically I'd say they're all capable of making a horrible room "workable" - but only Dirac is relatively transparent... in that it's much more subtle and not trying to completely restructure everything!

Bottom line for me... DSP for the win regardless - but well worth it to get something you can upload custom filters to or that comes with Dirac built in (provded it actually works that is). And yes, I should have highlighted the RC curve not the uncorrected in my charts... sue me.

 

[Pio2001]

I rather think that you forgot to turn on Dirac during the measurement.
The green and black curves are just the same, with different averaging.

I tried several room corrections with the Neuman KH-120. Since their frequency response is dead flat on axis, they obviously spoiled it. Some of the corrections changed the overall bass / treble balance a lot, trying to reach a generic target curve that was very different from the real one (defined by the natural response of the speakers in the room, minus the SBIR and room modes, that causes the localized peaks and nulls).

For me, this is the major problem of all full-band corrections : they target a curve that is not the good one. The good curve is specific of the speaker (because of its directivity characteristics), listening distance (near-field vs far field), and room (reverberation time).
It is much better to have the speakers listening window (average of frequency responses close to the tweeter's axis) flat, and then deal with room modes (equalization below 150 Hz) and sbir (eq from 150 to 600 Hz) separately. That's the only way to have something that sounds realistic. Above 500 Hz, our auditory system hears "through the room" and expect the speaker to have a flat frequency response in itself, not a flat frequency response measured from the listening point. The later is what is measured, but it is not what we hear. Our ear performs part of what Klippel calls "sound field separation". It does it partly by "gating" impulses present in the music, exactly like REW when we perform pseudo-anechoic measurements. The "gating" in our ears is called psycho-acoustic temporal masking (https://en.wikipedia.org/wiki/Auditory_masking#Temporal_masking), and lossy codecs such as mp3 take advantage of it.

 

[digicidal]

I rather think that you forgot to turn on Dirac during the measurement.
The green and black curves are just the same, with different averaging.

You know, I think you're quite likely right on that... I'll check again. I ran the calibration for Dirac when I had two plugins active, and although I know I checked - I may have posted the filter to the standalone plugin in Foobar rather than the plugin in Equalizer APO. I will attempt again tomorrow and then post updated measurement for it.

That's why it sounded more different than the chart looked like I'm sure (since foobar had it always on) - but I can't get it to post my saved project to the other control now... says the device isn't sync'd properly. So I guess I'll have to run that over from scratch maybe to get it to post.

EDIT: I've edited my original post with some explanations... no Dirac for some time is the TL;DR. Either I've done something colossally stupid in my configuration, or using Dirac VST's isn't totally ready for primetime in this use case (the former most likely I'd guess). In any case, I'll revisit in a month or so after resolving the issues and doing more research. In the meantime, Dirac works great - but only for Foobar it seems... everything else doesn't take the filter - including REW... so no way to comparatively measure it accurately without way more work than I'm willing to provide at the moment.

 

[North_Sky]

I like your avatar Markus, and looking forward to advanced analysis in the world of room correction systems.

 

[Head_Unit]

OK I'll add in here

But I'd hope the thread title might be changed to
"Effects of room correction systems (Audyssey, Dirac, YPAO, MCACC, AccuEQ, etc)"
in order to get more posts, and be more searchable.

Buchardt A500 without (red) and with (green) room correction in a medium size apartment. Holy cow!

 

[ernestcarl]

OK I'll add in here

But I'd hope the thread title might be changed to
"Effects of room correction systems (Audyssey, Dirac, YPAO, MCACC, AccuEQ, etc)"
in order to get more posts, and be more searchable.

Buchardt S400 without (red) and with (green) room correction in a medium size apartment. Holy cow!

With any kind of drastic 'correction', I'd still want to see detailed before and after distortion graphs of 90% to 100% at your set max volume.

 

[Head_Unit]

With any kind of drastic 'correction', I'd still want to see detailed before and after distortion graphs of 90% to 100% at your set max volume.

Do you mean out of concern for clipping? Or what do you mean by the 90%-100%? (Certainly my friend's Buchardt is likely getting a lot of boost, though he is in a situation where he can't crank it much, hence my recommendation for their calibrated loudness function).

 

[ernestcarl]

Do you mean out of concern for clipping? Or what do you mean by the 90%-100%? (Certainly my friend's Buchardt is likely getting a lot of boost, though he is in a situation where he can't crank it much, hence my recommendation for their calibrated loudness function).

Well, not just to potentially clipping effects. But what effect does any gross EQ correction does to the speaker's native distortion values. Does your EQ reduce it or signficantly increase it... important if you like to listen at loud levels or want to -- on ocassions -- max your speaker at your set full volume.

 

[ernestcarl]

Just an example:

All of my PEQ filters for this setup requires no additional headroom. Avoiding clipping and keeping higher order distortion harmonics in check allows me to listen at loud volumes without fatigue. It also minimizes boominess due to aggressive EQ boosts elsewhere in the room or in adjacent seats.

 

[RayDunzl]

But what effect does any gross EQ correction does to the speaker's native distortion values. Does your EQ reduce it or signficantly increase it...

My limited experience with digital room correction via AcourateDRC's filters applied to a miniDSP OpenDRC-DI reveals no unusual changes in the distortion profile.

Red - no DRC
Black - with DRC

Red - An uncorrected dip in the fundamental without DRC around 50Hz causes the harmonics, which are not attenuated by the same mechanism (in-room phase cancellation) in this case, to read as "higher distortion".

 

[ernestcarl]

My limited experience with digital room correction via AcourateDRC's filters applied to a miniDSP OpenDRC-DI reveals no unusual changes in the distortion profile.

Red - no DRC
Black - with DRC

View attachment 79468

Red - An uncorrected dip in the fundamental without DRC around 50Hz causes the harmonics, which are not attenuated by the same mechanism (in-room phase cancellation) in this case, to read as "higher distortion".

I think Dirac, Acourate and other smarter DRC's make more conservative boosts. With anything that you can do or modify manually, it's incredibly easy to overdo/overfill nulls -- I'm guilty of this myself. The 'bad' effect may not be obvious at 75 or 85dB SPL in the distortion graphs... but at 90 to +100dB SPL, there you might see some nastiness that weren't there before with no EQ boosts applied.