Redefine your budget room EQ 'flat' target curve to Harman's pro curve

[CherylJosie]

? Audyssey XT32 Versus Dirac Live | Stereo Listening Comparison

* TEST (Dirac Validation): [POST]275750[/POST]

Thanks for those links Bob. I read the first and am working on the second.

The two Audyssey curves, flat and reference, are both chewed up messes with substantially reduced subwoofer level. The authors proceed to fix the low frequency by manually adjusting the bass level as we have been suggesting here. THey then proceed to test for "imaging" and some such thing and declare the two systems sounding the same....Good news is they clearly show why Audyssey doesn't perform well in the way 99% of the users utilize it per above measurements. And their own subjective assessment prior to manual intervention that bass response was not correct.

Actually they modified the Dirac target to match the general shape of the Audyssey in-room EQ result and then compared Audyssey vs. Dirac to see if there was any audible difference in sound stage and imaging.

They apparently wanted to find out if the mixed phase filters with their better impulse response actually helped anything.

They insisted that measurements only at the sweet spot were "better", and that is what they proceeded to do.

Not quite. They used a single mic position and never moved the mic throughout the testing to control the number of variables. They were more interested in sound stage and imaging than they were in EQ and what they wanted to know was if the mixed phase filters of Dirac did anything to help with that. They concluded that any mixed phase improvements were inaudible.

I am surprised by that shelf just the same. I wish they had shown the uncorrected graph so we knew what it started with.

There is an uncorrected response plot in the thread at Home Theater Shack but it was plotted with a different tool apparently. It appears to show a broad depressed shelf in the bass just like the Audyssey resulting EQ does.

Note that the subwoofer level is set by the AVR based upon the measurement results that Audyssey reports and it never acutally measures the resulting EQ with sub integrated. This is an interface issue between the AVR and Audyssey, not an Audyssey error.

The dip at 70Hz could also be due to phase issues from similar mechanism.

The reason Dirac shows no shelf or dip like that might be if it EQs after the receiver has already set the levels.

They should have fixed this problem in the levels and phase immediately after the setup, to optimize the initial cal and avoid any confusion later. Oh well.

 

[Fitzcaraldo215]

I

I am surprised by that shelf just the same. I wish they had shown the uncorrected graph so we knew what it started with.


Sorry I was not clear. By 99% I meant the 1% who use the Pro kit and can create their own target curve. Otherwise, they will get what is handed to them (99% of the time) which in opinion, has always damaged the mid-range with that BBC curve, and anemic bass. I have tested it on many AVRs in a number of rooms and while occasionally I like it at first, when I listen some more I have to defeat it. But that is my experience. I am not saying at all that 99% of people are like me and just say otherwise .


We come from different places. I come from having experienced many other systems that work better to Audyssey, including manual optimization. Vast majority of people experience Audyssey as their first Auto EQ. By setting levels and delays right alone, there can be significant improvement which is what it does better than manual systems. It also gets rid of some low frequency resonances and that is good too. So it is not that it does nothing good. It is that it is not a complete, performant package so to me it always sounds like a step down.

I am surprised by that shelf just the same. I wish they had shown the uncorrected graph so we knew what it started with.


Sorry I was not clear. By 99% I meant the 1% who use the Pro kit and can create their own target curve. Otherwise, they will get what is handed to them (99% of the time) which in opinion, has always damaged the mid-range with that BBC curve, and anemic bass. I have tested it on many AVRs in a number of rooms and while occasionally I like it at first, when I listen some more I have to defeat it. But that is my experience. I am not saying at all that 99% of people are like me and just say otherwise .


We come from different places. I come from having experienced many other systems that work better to Audyssey, including manual optimization. Vast majority of people experience Audyssey as their first Auto EQ. By setting levels and delays right alone, there can be significant improvement which is what it does better than manual systems. It also gets rid of some low frequency resonances and that is good too. So it is not that it does nothing good. It is that it is not a complete, performant package so to me it always sounds like a step down.

We do not disagree by much about Audyssey, but I do not think it as bad as you overall. But, then I have not heard it as much in comparison to other tools you have used, some of them quite deluxe.

I think we agree their biggest problem is the BBC dip, which I always eliminated once I got the Pro kit. I did not find Audyssey's bass to be anemic. The graphs we are all referring to from HT Shack misrepresent that, I feel. There are countless independent REW, etc. user measurements in Audyssey forums that support my view.

The flat bass response in their target might not be desirable to many. From my personal experience and other objective data I have seen, I believe Audyssey does generally achieve flat bass response per the target and which is considerably smoother than no EQ. I think Audyssey proceeded on several assumptions. They wanted a flat target in the bass at movie "reference level", then their Dynamic EQ, if used, would proportionally and dynamically provide the necessary Fletcher-Munson type bass lift at lower volume settings. It is an intriguing idea, actually. A single target curve is not necessarily correct for all volume settings. Unfortunately, this scheme does not work with music where there is no accepted or de facto reference level.

So, listening volume level is another of those infernal variables affecting the choice of a proper target curve or ideally a family of curves for different listening volumes.

But, yes, overall, Audyssey has become the mass market Hyundai of EQ systems and it seems content to stay in that market segment. I still believe that many audiophiles are better off using it than not, if they have no other choice, even if it does not provide the ultimate. And, they have a choice. If they do not like what it does, they can turn it off. It introduced me a quite a few other audiophiles in my circle to better sound via EQ, as the results were sonically quite positive and quite obvious. Like me, they have all moved on to something better.

 

[amirm]

I think we agree their biggest problem is the BBC dip, which I always eliminated once I got the Pro kit. I did not find Audyssey's bass to be anemic. The graphs we are all referring to from HT Shack misrepresent that, I feel. There are countless independent REW, etc. user measurements in Audyssey forums that support my view.

I have yet to see frequency response that is different. From the AES double blind test of it, these are the results: http://www.madronadigital.com/Library/Room Equalization/Room Equalization.html

Its response is the one in teal at the bottom. We can see that it even dialed out the room gain that was there. Here are the relative listening results from the trained panel:

Ideal response would be flat at zero line. We see that RC6 which is Audyssey did the worst in bass and mid-range. The mid-range is due to BBC dip which took the B&W 802 speakers which already had a problem there, and made it doubly worse. And the bass due to it making it more anemic than doing nothing. The correlation with the measurements is quite high.

The flat bass response in their target might not be desirable to many. From my personal experience and other objective data I have seen, I believe Audyssey does generally achieve flat bass response per the target and which is considerably smoother than no EQ.

No disagreement.

I think Audyssey proceeded on several assumptions. They wanted a flat target in the bass at movie "reference level", then their Dynamic EQ, if used, would proportionally and dynamically provide the necessary Fletcher-Munson type bass lift at lower volume settings. It is an intriguing idea, actually. A single target curve is not necessarily correct for all volume settings. Unfortunately, this scheme does not work with music where there is no accepted or de facto reference level.

Dynamic EQ came after Audyssey so clearly they did not have that in mind when designing it. I think this is classic inexperience in the field. The origin of the technology is from academia where they assumed flat was desirable, not having decades of experience that others have in acoustics that says otherwise. Difference between theory and practice.

I also don't think Dynamic EQ was designed to fix Audyssey. This is just the modern day Loudness button that we used to have in 1970s. It was yet another feature to sell and advertise.

So, listening volume level is another of those infernal variables affecting the choice of a proper target curve or ideally a family of curves for different listening volumes.

I hear this argument often but there is a reason we did away with Loudness button. It does not matter that much in practice. Most of us listen to content at a narrow range of volumes. Within that, one target curve works very well. Dynamically changing the frequency response based on population averages reflected in Fletcher-Munson again sounds good on paper, but in reality I don't think it is necessary or useful.

The reason to have multiple targets and be able to switch between them would be the content you play. Sometimes you may not want your sub whaling away.

But, yes, overall, Audyssey has become the mass market Hyundai of EQ systems and it seems content to stay in that market segment. I still believe that many audiophiles are better off using it than not, if they have no other choice, even if it does not provide the ultimate. And, they have a choice. If they do not like what it does, they can turn it off. It introduced me a quite a few other audiophiles in my circle to better sound via EQ, as the results were sonically quite positive and quite obvious. Like me, they have all moved on to something better.

I have it in multiple AVRs and as I have mentioned, I leave it off. I think we are going to introduce people to an Auto EQ system, it needs to be more performant than this.

By the way, Audyssey also downsamples everything to 44/48 Khz as implemented in AVRs. So if you use it, forget about any high-resolution audio passing through.

 

[CherylJosie]

The mid-range is due to BBC dip which took the B&W 802 speakers which already had a problem there, and made it doubly worse.

Are you sure about that?



It appears that Audyssey significantly narrowed the dip by about half an octave and shifted it down about 1000Hz too, although the response also has residual peaks and dips all over it including two apparent peaks on either side that seem to increase the apparent magnitude and bandwidth of the dip.

The lesson here might be that deliberate peaks and dips in response are subject to the risk that some residual error will align with and exaggerate such tweaks if the algorithm has loose error limits (probably due to lack of correction power). Per your KISS comment to me, it would seem that overly complicating a rough performing algorithm is a bad idea.



What I am seeing in the plots is that Audyssey fails to extend the frequency response (they claim that they will not correct past the -3dB point) and the resultant response is also ragged. In some senses it seems to be just as ragged as the speaker itself if not more so.

I am seeing as much evidence that the problem is in the result as in the target.

Sean's own conclusion is that smooth and extended response was more preferred in general and that is in agreement with long-standing commonly accepted belief.

If Audyssey had more processing power and tighter error limits and corrected out to ~-6dB point it might have performed much better in the study. Of course, extending the response beyond the limitations of the speaker also puts much greater strain on them.

That might work out in the pro world, but is not necessarily a good idea with budget-oriented equipment. 3dB is double the power; 6dB is quadruple.

I am also wondering how the current MultEQ-XT32 performs compared to the one in the Harman study that was of a much older 'Pro' version. Forming conclusions about the later versions based on this old version might cause mistaken impressions. I do not find my system to be 'colored' by ragged response or lacking in bass extension with Audyssey engaged, but I do hear the percieved brightness in the on-axis response of a system corrected to flat power response.

The one thing that the Harman listeners really seemed to object to (other than the lack of bass extension) can maybe be explained by this:
https://audyssey.zendesk.com/entries/94162-multeq-target-curves

Chris,
Where would I find a graphic representation of the "standard" Audyssey target EQ curve.

Hi Rob,

It's not generally something that we publish. The curve is flat out to about 8 kHz, then dips to about –2 dB at 10 kHz and then a little more at 20 kHz...(emphasis added)

Combined with a ragged response lacking in bass extension and adding a midrange dip, rolling off just the top end of the treble seems like a bad thing to do. From the measured response, it would seem that Audyssey not only failed to extend the treble beyond their measured -3dB point, it also attenuated it even more.

I hear this argument often but there is a reason we did away with Loudness button. It does not matter that much in practice. Most of us listen to content at a narrow range of volumes.

Without loudness compensation and dynamic range control, movies would be unlistenable in my apartment at levels that will not disturb my neighbors, particularly at night.

I am very glad that we have not 'did away with' the loudness button. It has its uses.

I have it in multiple AVRs and as I have mentioned, I leave it off. I think we are going to introduce people to an Auto EQ system, it needs to be more performant than this.

Who is willing to add $700+ to the cost of a receiver for a more sophisticated EQ when most people buy the $500 model and have no clue how to plug it in without asking their teenagers to help them out with it?

This objection also applies doubly when considering that correction with the capability to significantly extend the frequency response beyond the inherent limitations of the speakers necessitates more power and/or efficiency. I think many people in the pro world lose sight of the financial implications of a 'sky is the limit' mindset that most people find wasteful if not repugnant.

I have found that the 2EQ version of Audyssey that comes in that $500 receiver does so little correction as to be worthless and its effect is nearly inaudible to me. Most people would probably not even notice if it was on or off but in such receivers it must be on to get the benefit of DEQ and DV. Your comment seems more apropo if aimed at that level IMO since a little more processing power and a little more appropriate target response could do a lot to improve such system.

 

[amirm]

Are you sure about that?

Yes. Here is the report itself: "All of the room corrections except RC6 [Audyssey] were able to correct the wide 2 kHz dip in the sound power response of the loudspeaker (see RC4)."

It appears that Audyssey significantly narrowed the dip by about half an octave and shifted it down about 1000Hz too, although the response also has residual peaks and dips all over it including two apparent peaks on either side that seem to increase the apparent magnitude and bandwidth of the dip.

If you look at the B&W's in-room response, its droop was wider but the right side was only up 2-3 dB. Audyssey made that much worse by creating a 5 dB difference. 5 db response variation at such a sensitive frequency is a big deal. I have measured Audyssey and in every case it has done the same thing by accentuating the depth of the dip.

 

[amirm]

Here are a tally of observations from the expert panel. Again, RC6 is Audyssey.

 

[CherylJosie]

Yes. Here is the report itself: "All of the room corrections except RC6 [Audyssey] were able to correct the wide 2 kHz dip in the sound power response of the loudspeaker (see RC4)."

This is sort of a 'when did you stop beating your wife' kind of statement. Audyssey target in this study did not attempt to 'correct the wide 2kHz dip in the sound power response of the loudspeaker'. It attempted to superimpose a dip of its own design so a dip is expected. My opinion is that such dip might help in some cases but a well-designed speaker is going to have a smooth power response by design (this one did not so I keep wondering why the B&W is so popular).

More on this in a second... stay tuned.

If you look at the B&W's in-room response, its droop was wider but the right side was only up 2-3 dB. Audyssey made that much worse by creating a 5 dB difference. 5 db response variation at such a sensitive frequency is a big deal. I have measured Audyssey and in every case it has done the same thing by accentuating the depth of the dip.

OK so now we get to the heart of the matter in this question. How does a correction algorithm function on a fundamental level and what are the implications for using one on a speaker?

In my opinion, all of the products on the market are wrong by design. Based on what broad top-level info on psychoacoustics, physics, and electronics I have gleaned from the debate online (plus my electrical engineering background), I think that a fairly comprehensive top-level design can be constructed that addresses the fundamentals of a speaker in a room without imposing any preconcieved notions of 'correctness' on the target. Admittedly I am not expert in anything audio so the picture may not be perfect but I think the outline is fairly obvious if one is looking for it.

 

[CherylJosie]

Well that was interesting. The anechoic response is so good the graphic did not even show up. Hm. Try, try again...



OK so now it is a ghostly image. Check the Harman study if you want their actual image.

Speakers are designed with the help of anechoic chambers. The reason is that in an anechoic chamber it is possible to measure the speaker directly without interference from the environment on its performance.



Here is the B&W anechoic response, or a facsimile thereof. The presentation does not specify how it was actually measured.



This is the JBL subwoofer response.



Subwoofers are usually measured outside because the bandwidth of absorption even in an anechoic chamber does nothing about the cabin gain. The pressure in an enclosed space adds a sort of 'DC term' to the response curve that no absorption, diffusion, or resonance can remove. Anechoic large enough to test subwoofer performance is prohibitively large.

One factor in such groundplane measurement is the doubling in power from 'half space'. All the energy radiated toward the ground is reflected back upward.

So how do we reconcile all of this information? The performance of each part is measured and designed differently, and there are more than one way to measure. Full range speakers are measured on-axis and off-axis because they are directional radiators at some frequency. Subwoofers radiate largely omnidirectionally so they are only measured from a single direction. Does this information matter?

It turns out that the information matters greatly. The on-axis response of a speaker is designed to be flat and extended because that is an easy repeatable reference target to specify, implement and hit, and the off-axis response is designed to be smoothly attenuating in the treble as the directivity increases. The degree of directivity is a design choice on the part of the speaker designer and it affects the way the speaker performs but is a separate independent variable in the equation that only the speaker knows for certain.

 

[CherylJosie]

Harman measured its room EQ in an echoic room of course because that is the typical listening environment.



Here is the 1 seat performance of all the EQ plus the speaker without EQ is measured too in black trace. The equalization that was most preferred was the equalization that most resembles a smoothed and extended version of the unequalized speaker response.

Disregard the second (green) trace for now. That is a 6 seat EQ shown on a 1 seat measurement.



Here is the 6 seat performance of all the EQ plus the speaker without EQ is measured too in black trace. Again, the equalization that was most preferred was the equalization that most resembles a smoothed and extended version of the unequalized speaker response.

Disregard the first (red) trace for now. That is a 1 seat EQ on a 6 seat measurement.

The two most preferred cases were the Harman 1 seat EQ in the sweet spot (red trace on first plot) and the Harman 6 seat EQ off-axis (green trace on the second plot). In each case, the algorithm uses an extraordinary amount of processing power and very tight limits to optimize the tuning for that specific case, and it very closely resembles the unequalized speaker response with smoothing and extension. The average slope of the target very closely matches the speaker itself.

This points out that humans hear many aspects of the speaker simultaneously and are capable of distinguishing them. The smooth anechoic on-axis response is audible, as is the treble-attenuated and 'dippy' off-axis response. The speaker and room combined create a somewhat acoustically transparent transfer function that melds the two aspects together into a single perceived speaker image and associated reflections in the human brain. We see an 'image' of the speaker that is distorted but still a valid representation of the actual design target because speakers are designed to sound good in rooms.

So let's investigate the 6-seat EQ with a 1-seat measurement and the 1-seat EQ with a 6-seat measurement.

The green trace in the top plot and the red trace in the bottom plot demonstrate that the listening position strongly affects the actual response of the speaker in a room. It is impossible to hold them both to the same degree of smoothness when listened to away from where they were intended to be listened to, regardless of how much processing power is thrown at them.

What these traces do not show is that the selection of the speaker, the placement in the room and yes even the choice of room also strongly affects the actual response of the speaker in a room.

Now what happens when we apply a target that models itself after the actual in-room response of the speaker is that a compromise is struck between the on-axis response and the off-axis response. All the radiated energy is roughly integrated (infinitesimally summed as a function of radiation angle) at the listening position with the added characteristics of the room superimposed, such as cabin gain, boundary gain, modes, reflections, reverberations etc. The angle-variant transfer function for each direction of radiation adds a distortion to the summation, with some aspects of that distortion readily addressed by EQ and others not.

Since we are humans we can do some of the processing ourselves. Fortunately the part we suck at (bass response) is readily addressed by EQ whereas the part that room EQ sucks at (reflections and reverberation) is where humans are much better at it.

Room modes can be relatively easily addressed with smoothing. Reflections not so much. That is why some people prefer to just EQ the subwoofer.

With EQ that uniformly applies a fixed target curve that applies the same curve on all speakers in the system regardless of type or placement or room, I can see why EQ over the whole frequency range or over all the speakers might be less than optimal. Tarring all speakers with the same EQ brush is fundamentally flawed approach. It is better to apply EQ to the one speaker in the system that is both isolated and strongly affected by room modes than to contaminate all the speakers with a target that does not resemble a smoothed and extended version of the actual unequalized response of the individual speaker in its specific location within a given room.

The red trace on the 1-seat measurement and the green trace on the 6 seat measurement clearly demonstrate that even in the best case room correction has fundamental limits that impose a tradeoff between listening position and result that an algorithm cannot overcome with processing power.

So where does this leave us? To me, the study leaves many unanswered questions, but I will hazard some guesses.

 

[amirm]

With EQ that uniformly applies a fixed target curve that applies the same curve on all speakers in the system regardless of type or placement or room, I can see why EQ over the whole frequency range or over all the speakers might be less than optimal.

All auto EQ systems have built-in limits to deal with what the speaker may not be capable of doing. They don't always do it right and vary on amount of information they have. Harman's ARCOS system is the only one that knows exactly what your speakers are designed to do since it is only provided as part of their JBL Synthesis package. The first task is telling it the speaker model number. Other systems like Audyssey as you know limit the db correction in low frequencies as to avoid damaging equipment.

That aside, the nature of target curve is an approximate overlay. You should not analyze it in the specific. Proper correction is actually more involved than what you see. Harman's system applies a psychoacoustically aware filtering to the measured response proportionally to the frequency so that it doesn't see the ups and downs that you can't hear. It is this stepping back from the tree to see the forest that makes their correction great starting point for any system regardless of speaker type. Just because we don't know the exact target, doesn't mean we choose flat or some other curve that we know to be wrong. We perform enough listening test to find the curve that generates most positive results. This is what Harman has done.

Tarring all speakers with the same EQ brush is fundamentally flawed approach. It is better to apply EQ to the one speaker in the system that is both isolated and strongly affected by room modes than to contaminate all the speakers with a target that does not resemble a smoothed and extended version of the actual unequalized response of the individual speaker in its specific location within a given room.

A target curve is a high-level strategy, not micro level optimization of a speaker in a room. It is designed to bring order to a rather chaotic situation. Experience again shows it be quite effective.

The red trace on the 1-seat measurement and the green trace on the 6 seat measurement clearly demonstrate that even in the best case room correction has fundamental limits that impose a tradeoff between listening position and result that an algorithm cannot overcome with processing power.

There is although the problem can be significantly made better using multiple-subwoofers, and computer optimization of delay, level and filtering of each. See http://www.madronadigital.com/Library/Computer Optimization of Acoustics.html

 

[CherylJosie]

My experiments with the equalizer in the feedback path demonstrated that the cutoff point (-3dB in the case of Audyssey) that the algorithms in the receiver depend upon to tell them where to stop correcting the response are directly affected by the overall slope of the response that the algorithm is hearing.

http://www.aespeakers.com/phpbb2/viewtopic.php?t=518


But what we already know is that the slope is dependent on the cabin gain. So any fixed target slope is by definition wrong because it will redefine a native aspect of the speaker response that cannot be corrected by EQ.

Personally I am no fan of cabin gain or any other room artifact. I prefer to blast my system with all the doors and windows open regardless of whether I am in a car or at home. It just sounds cleaner to me.

Yet I can understand that in any given environment that fits the definition of a 'small' room, there will be cabin gain, so I prefer an algorithm that tries to measure the actual power response and then deliberately ignores cabin gain and dispersion characteristics that a fixed curve targets, but just listens to the speaker and asks it where its happy operating point is.



A high-performance system can benefit from EQ that extends the bass response and flattens the entire response because it has acoustic power capability to spare (red trace). A budget system, not so much. There is a compromise between price and performance and every person has to make an individual decision on how much EQ they are going to buy because it is not free.

 

[CherylJosie]

Now there are other aspects in this equation that also need considering, so let's get on to that part.



Cabin gain within a single room is basically a fixed quantity so the slope of the curve is going to be a function of the room size more than anything else.

Placement is a function of the channel definition so optimization placement for sound has limited possibilities. Placement will strongly affect the rest of the variables such as modes, reflections, and reverberations. Of those aspects, modes are easily addressed by smoothing. Room treatments can also help.

Speaker choice is essentially arbitrary but the typical practice is to use capable speakers for the l/r/sub and smaller speakers that are more easily placed in tight spaces for the rest. The summation of total energy at a given listening position will strongly reflect the choice of speaker, particularly how smooth and directional its response is.

Harman totally ignored these considerations in their study. They chose to emphasize only a single factor.



Since listeners are less discriminating when listening to multichannel system, Harman chose a single speaker (mono) and neglected all other relevant considerations.

One such consideration is the cost of the system. Since listeners are less discriminating in multichannel, it seems a brain-dead assumption that room EQ is less of a driving factor for such applications. An expensive EQ is probably not required in many cases and may not even be desired if such choice prevents a person (like me) from experimenting with 11 channels.

Onkyo has already abandoned the fixed (therefore wrong) and proprietary (and also apparently too expensive) target curves of Audyssey and attempted another solution. They measure the response of the (assumed best selected and placed) front l/r speakers and impose that measured response, or some facsimile of it, upon the entire system. This is also wrong, but potentially better, since the best speakers in the system in the best placement have the best chance of enabling accurate measurement of the cabin gain.

It is also probably appropriate to forego room EQ on the best speakers in the system as a first-order approximation of not corrupting the entire system with a fixed target. Still, it would have been nice if they at least attempted to smooth out the response at least in the bass where room modes can be ameliorated.

Unfortunately this AccuEQ approach of Onkyo seems to employ a terrible quality of algorithm (per the one on-line measurement I have seen) and also ignores the rest of the considerations about choice of speaker, its directivity, its placement etc.

Imposing the power response of one speaker in one placement upon a dissimilar model in a different placement also seems a flawed solution.

A more appropriate solution might be to ask the user to identify all the speakers of the same model(s) into a group so that the algorithm can compare their individual responses, apply some averaging, and get a decent picture of how a speaker like that behaves in the given environment. Then each of the speakers in a group can be individually EQd to the averaged target that was derived off their unequalized response.

The front l/r might be one group, the 4 surrounds a second group, the center a third group, and the wide/height a fourth group, with the subwoofers of course the fifth group. That is how my system is configured anyway.



The smoothing function is up for debate, but the B&K target seems to be a worthy contender.

It seemed to help my system when I applied an attenuated version of its inverse frequency response to my microphone signal, despite the corruption of the flat-targetting algorithm by the alteration of the measured -3dB points that it used to detect where to stop correcting and despite the apparent difficulty the algorithm had adapting the EQ to such debased input.

The B&K curve has the advantage that it does not put such high demands upon the acoustic power capability of the system as Harman's straight line will. It might also model the actual in-room response better since every speaker I ever met on the Internet had an inherent inverted bowl shape to its unequalized in-room response.

So given an inverted bowl-shaped target, the appropriate curve to fit the measured response would be second-order, and least squares is probably about as appropriate as any error function to apply since it seems to be good for convergence.

A selector switch to instantaneously compare straight-line to bowl-shaped would be a nice feature to have, just in case one or the other turned out to be preferable in any given circumstance.

If I had a million dollars to spend on this project, that is how I would EQ my system. I would probably need to design the EQ myself since none of the options available on the market right now seem to address all the relevant aspects IMO.

The final tweak on all of this would be a knob that could be used to skew the target slope of each speaker, just in case something went weird in the measurement process that corrupted the cabin gain estimate and the wrong target was chosen.

For a second-order fit, there could be a second knob to adjust the concavity in case that was detected improperly. The bowl-shaped feature translates into an estimation of the switch between omnidirectional radiator and directional radiator as a function of frequency and the Schroeder transition frequency enters into that mix also by muddying up the picture with sparse room modes below about 300Hz.

The amount of correction the algorithm attempts (as reflected in its error limits) could get wider below Shroeder as an implicit recognition that the large deviations due to modes will be harder to smooth out. That might help reduce the complexity and cost of the algorithm. Audyssey along with probably most if not all EQ already employs variable bandwidth of the correction bins to save processing power, so frequency-dependent and bandwidth-dependent error limits seems like it might be another method to address cost without sacrificing too much performance, or maybe might even enhance performance. Room EQ already ignores deep narrow cancellation nulls since they are basically uncorrectable.

I guess what I am stating here is that I reject the idea of a fixed target to EQ a speaker. My preference is to let the speaker and its placement within a room directly tweak the target toward its measured power response with some smoothing and extension superimposed. Then the native sound of the speaker in a room is preserved to the greatest extent possible along with the original flat on-axis anechoic intent of the speaker designer. The compromise is applied right where it belongs by the nature of the beast.

I guess for larger venues, some nearfield measurement of the system might be appropriate also since air tends to attenuate treble and the algorithm should be made aware of that factor too. The only way I can think of to do that is to take a specially identified and defined measurement for the algorithm to compare near and far response so it can estimate how much the air is affecting the sound and apply appropriate compensation that averages the treble response over the listening area toward the native on-axis nearfield response.

OK well I am heading off to the REW path now to work on my room treatments. When I have some measurements I will put everything into the new thread.

 

[CherylJosie]

Oh yes, sorry I forgot to include one last point.

The 3dB limits of correction need to be determined by how well the actual in-room response fits to the target. A bowl-shaped target not only allows for less power requirement to achieve the end result, but also allows for more extension in the bass since the bowl fits the inherent response better. This means that correction can be applied to lower frequency without requiring so much acoustic power or correction power for a potentially superior result overall within the same limitations.

 

[Fitzcaraldo215]

Yes. Here is the report itself: "All of the room corrections except RC6 [Audyssey] were able to correct the wide 2 kHz dip in the sound power response of the loudspeaker (see RC4)."


If you look at the B&W's in-room response, its droop was wider but the right side was only up 2-3 dB. Audyssey made that much worse by creating a 5 dB difference. 5 db response variation at such a sensitive frequency is a big deal. I have measured Audyssey and in every case it has done the same thing by accentuating the depth of the dip.

Amir, I agree. As I have been saying all along, Audyssey's biggest problem, among others, is the BBC dip, which if memory serves is -6dB on the target curve. But, note that this should not "double up" on that issue with B&Ws, as you stated earlier, unless you mean it has made response worse than it was in this region. In which case, yeah, I totally agree.

Although many other EQs wil only trim response peaks and leave troughs alone, I believe from experience that Audyssey will also fill response troughs - with built in limits to prevent damage. In any case, even given the B&W native trough behavior in that frequency region, the result would just be closer to the Audyssey target curve rather than doubling up the B&W dip plus the Audyssey dip. The Audyssey curve has merely superseded the B&W curve. But, also agreed Audyssey appears to be pushing the B&Ws down further in that notch than they do by themselves because of the infernal Audyssey BBC dip target.

All right, all right, I get it. You do not like Audyssey, no way, no how. But, if that is all I had, I would still use it, unlike you. I suspect millions of mass market AVR users agree, for better or for worse. They just have not had the privilege of hearing the better EQ systems you have, and they could not afford them anyway.

 

[CherylJosie]

Harman's ARCOS system is the only one that knows exactly what your speakers are designed to do since it is only provided as part of their JBL Synthesis package. The first task is telling it the speaker model number.

Do you have a link to the user manual? I would like to see what it says about manual entry of room parameters, curve fit to the given room, etc.

We perform enough listening test to find the curve that generates most positive results. This is what Harman has done.

I am not seeing evidence of that. Harman defined 'most positive results' as being what pleases their trained listeners from among the choices they were offered. They never asked their trained listeners to tweak the curve to taste so we will never know if any additional improvements were possible.

They applied a prodigious amount of processing power to a system with prodigiuous amounts of acoustic power to correct to an arbitrarily straight-line response, without bothering to even attempt adding any contour to their solution. I would personally have liked to see some contour in the target and comparison of preference to see if even greater extension and perceived neutrality were possible, but this was not even attempted.

According to someone in another forum who claimed to know, that target was actually based upon prior listening tests of that single speaker in that single room with that single placement while playing those actual selections of material and that is a customized (as opposed to generalized) advantage that none of the other EQ was provided with. If you have any information disproving this claim, please provide it, because it is the only insight I could get into the inner workings of Harman's target. You are the only one I have seen claiming that the algorithm chooses the target based upon generalized psychoacoustics or any other criteria.

They completely ignored the B&K or second-order solution in their own design while including only one competitor for comparison (Audyssey) that seems to attempt such contour, and I have already pointed out the shortcomings of such EQ (limited extension, rolling off just the high treble, another arbitrarily predefined target, inaccurate results etc.) This is not a fair test of what order approximation is best. It put a behemoth system up against a relatively puny and twisted competitor IMO and then attempted to generalize the result inappropriately in the conclusion of the presentation.

The only variables were the listening position and the measurements. Even the choice of material was constrained to pop music. That does not equate to 'most positive results'. It equates to most positive results under their severely limited design and testing of the algorithm.

If you have any other evidence to show how Harman generalized this solution and tested it with trained listeners across multiple rooms with multiple speaker types and multiple speaker placements, please post a link. Otherwise your statement is not accurate.

A target curve is a high-level strategy, not micro level optimization of a speaker in a room. It is designed to bring order to a rather chaotic situation. Experience again shows it be quite effective.

Please demonstrate to me how simply smoothing and extending the native response of a speaker within a given room amounts to 'micro level optimization'. It seems to me to be the most natural and widespread practice based upon my experience. Remember your stereo graphic EQ? It has no predefined target and it is optimized per speaker using a brain that 'hears through the room'. A single power response target applied across a whole system obviously does not achieve the same result. Humans do not hear only the power response.

I think rather 'forcing a foot into the wrong shoe' is a more appropriate characterization of Harman's approach, along with all the other approaches that use a single fixed target across an entire system and do not use any input from the actual measured in-room response to tweak the target.

Remember, what I propose has never been attempted (unless someone else was bright enough to think of it first). I think it is unfair to pre-emptively dismiss it out of hand with vague references to psychoacoustics. What I have stated here is in total agreement with what Floyd Toole has stated, with the one exception that I think perhaps room EQ that fits the specifics of the individualized problem has a very good chance of improving overall sound without corrupting anything unacceptably, as demonstrated by prior experience you allude to.

I am just proposing to drop the obvious pretense that the laws of physics and psychoacoustics boil down to a straight line one-size-fits-all solution. Take a look out your window and tell me how many straight lines you see in nature and then explain how a straight line of a single predefined slope, or any other shape for that matter, accurately reflects the psychoacoustics of cabin gain, room modes, directivity, reflections, and reverberation for every speaker in a given system. Sherlock Holmes would tell you that when the impossible is eliminated, the answer becomes obvious. Highly nonlinear phenomena in multiple variables do not a generalized linear optimization make.



Take another look at this plot and explain to me why the psychoacoustics of the B&W speaker with JBL subwoofer in the Harman room just happens to generate EQ target (red trace) that looks very much like a straight-line approximation of the unequalized response of the system (black trace). Believe the evidence of your eyes, Amir.

People with plenty of experience have already commented on this thread that the room affects the target. It is a simple extension to claim that the speaker also affects the target and I suspect that we could easily develop experimental evidence confirming this too.

It all makes sense on a theoretical basis because the power response is highly dependent upon both the dispersion pattern and the room acoustics. The linear on-axis response as designed and measured in anechoic environment is the 'holy grail' target, but the power response is the only measurable quality available within any given room.

If we want to preserve the original intent of the speaker designer it is incumbent upon us to measure its actual power response in any given application and extract information about the intended response from that measurement or we are abandoning the original intent. Call me a 'strict constructionist' if you will, but that is my personal opinion and until someone provides evidence to the contrary I am sticking to my guns here.

I am just advocating for the room and the speaker to get a vote in the target, just like Harman allegedly did when they created their own EQ target based upon prior testing with the implemented sound system. I am advocating that Harman or whoever extend the same courtesy to every speaker in every room and every placement that they allegedly gave to the B&W with JBL subwoofer in their mono listening tests.

Harman is also guilty as charged per your 'micro level optimization' criteria if what I read about where their target came from is true.

Their chosen first-order (straight line) might indicate that they also ignored potentially important psychoacoustic aspects of dispersion pattern in that target too.

I am advocating for a potentially improved approach that takes advantage of more information without adding significant cost, not some sort of misguided rube goldberg overcomplication that explodes in your face with disastrous results at inflated price. We should at least characterize the debate in terms that fit the actual parameters under discussion and so far IMO you have not provided any explanation as to why this approach I am advocating would adversely affect anything at all.

Respectfully, as always, and thank you for co-founding your forum Amir. I really enjoy it here.

 

[amirm]

Placement is a function of the channel definition so optimization placement for sound has limited possibilities.

That room is custom designed listening space for this type of testing. Speakers are carefully located to minimize modes and interact properly with the acoustic treatments. Nothing about that is ad-hoc or left to chance.

Placement will strongly affect the rest of the variables such as modes, reflections, and reverberations. Of those aspects, modes are easily addressed by smoothing. Room treatments can also help.

All of which is carefully documented by Harman. The room as I said, is purpose built to be a) typical of home listening spaces as far as reverberation time and b) purpose built and optimized to sound good.

Speaker choice is essentially arbitrary but the typical practice is to use capable speakers for the l/r/sub and smaller speakers that are more easily placed in tight spaces for the rest. The summation of total energy at a given listening position will strongly reflect the choice of speaker, particularly how smooth and directional its response is.

Harman totally ignored these considerations in their study. They chose to emphasize only a single factor.

You mean they are not allowed to test Auto EQ systems with freestanding speakers augmented by a sub??? The EQ systems tested were high-end solutions at the time and they are routinely sold with powerful systems, not bookshelf speakers. They had to pick a system and they selected the highest volume high-end speaker brand there is (B&W) and added a sub to it. How would any other choice be defensible?

But yes, they did focus on one factor which is when given the exact condition to multiple systems, how they would perform. Each system scored differently, clearly differentiating those that worked well, and those that did not. Among those that did work well was also a non-Harman system so nothing was cooked so that only their system sounded good.

Since listeners are less discriminating when listening to multichannel system, Harman chose a single speaker (mono) and neglected all other relevant considerations.

Can you just give a lit of those considerations? I am having a hard time extracting them from the full text.

One such consideration is the cost of the system. Since listeners are less discriminating in multichannel, it seems a brain-dead assumption that room EQ is less of a driving factor for such applications. An expensive EQ is probably not required in many cases and may not even be desired if such choice prevents a person (like me) from experimenting with 11 channels.

Not at all. Content very frequently localizes to one channel such as voices in movies or stereo playback. While the euphonic effect of many channels can make one ignore more flaws, that is no excuse to leave all the channels uncorrected.

It is also probably appropriate to forego room EQ on the best speakers in the system as a first-order approximation of not corrupting the entire system with a fixed target. Still, it would have been nice if they at least attempted to smooth out the response at least in the bass where room modes can be ameliorated.

A performant Auto EQ does a lot things:

1. Uses multiple subs to get smooth frequency response across the full listening area.
2. Smoothes the frequency response
3. Makes sure the sub and mains crossover correctly as to create the response of one speaker. It is incredibly complex to accomplish blending one sub with multiple speakers in the room but that is what you need to do. Now add multiple subs to this and scope becomes infinite. You need computer optimization.
4. Target curve overlay

This is why I call this bucket computer optimization of acoustics as opposed to just Auto EQ. You can see all of these steps in the JBL Synthesis Arcos:

If I had a million dollars to spend on this project, that is how I would EQ my system. I would probably need to design the EQ myself since none of the options available on the market right now seem to address all the relevant aspects IMO.

If you had a million dollars, the place to start to solve this puzzle is not at EQ stage but computer modelling of the room for optimal design and placement of subwoofers. That is what we did by working with Keith Yates and fluid dynamic modelling to determine the number and location of subs. Here is one of the visualizations you get out of that:

On the left is a single sub in the corner which many people use. We see that it creates very high response variations spatially (locations) and in frequency domain. The triple sub helps greatly there although EQ is mandatory still. The cost of our theater is in $350,000 range as a way of reference. This part of the design costs around $10,000 to $15,000 depending the complexity of what you want modelled (everything from drywall to seating and people can be included). The acoustic treatment cost us $20,000 and we built some of it to get that. You can read more about that here:
http://www.madronadigital.com/Showroom/HomeTheater.html

 

[CherylJosie]

Here is one of the visualizations you get out of that:

That is one of the most interesting simulation I have seen. Great graphics.

I am unable to upload any more images for some reason so I basically cannot respond to the rest of your comments.

I would like you to know that I support the Harman study and its stated conclusions.

What I do not support is universally applying it or drawing sweeping conclusions about its obectives and results. It tested a specific set of pro systems. It was not a research project in the optimal EQ target or a validation of the optimal EQ method.

I know you have not stated any such thing but I am perceiving a lack of skepticism that implies such undercurrent.

What my statements were intended for is to float an idea for a more optimal solution that can be more universally applied to budget systems as well as pro systems.

The JBL EQ you referenced is going in the opposite direction. It is customized to specific speakers. In that sense it partially supports my contention that the EQ target should be based upon measurements of the actual system that is being used.

 

[steve williams]

Fascinating thread. Would you mind if I made it a sticky?

 

[amirm]

What I do not support is universally applying it or drawing sweeping conclusions about its obectives and results. It tested a specific set of pro systems. It was not a research project in the optimal EQ target or a validation of the optimal EQ method.

Well, we have five different target curves in the study so in that regard, there is a comparison of that many. And there was good separation of subjective results between two buckets: those that had used flat response versus those that had sloping down. The latter did better. Outside of that, I don't think more precision is fruitful. Ultimately what tonal curve we need for each and every piece of content is unknown because we don't know what the producers and talent heard in their rooms. This is why I continue to say that the Harman curve should be the starting point and you adjust based on taste.

Numerical analysis to determine something that ultimately has no calibration at the source is a fruitless endeavour.

The JBL EQ you referenced is going in the opposite direction. It is customized to specific speakers. In that sense it partially supports my contention that the EQ target should be based upon measurements of the actual system that is being used.

Since that information does not exist as a practical matter for anything other than Harman speakers, Auto EQ systems still need to be performant. Hence, when we find something that works for majority of people, we should use that. And the sloping down target curve is one of those. It has been shown over and over again to be the right strategic choice.

 

[CherylJosie]

Well, we have five different target curves in the study so in that regard, there is a comparison of that many. And there was good separation of subjective results between two buckets: those that had used flat response versus those that had sloping down. The latter did better.

Not exactly. Audyssey curve also sloped down and it did the worst. It may have been flat in the bass but it attenuated the treble.

It did worse than the curve that was flat all the way across. At least the other flat curve was deemed as good as no EQ.

There could have been other reasons why Audyssey performance was undesirable and probably the worst thing about it IMO is how ragged it came out, just as bad as the uncorrected response if not worse plus there is that BBC burble in the midrange too and the bass extension was pulled back by the flat part of the bass curve and the -3dB limit. There were so many things going on it is IMO invalid to point to any one thing and blame it for the poor performance in listening tests.

Outside of that, I don't think more precision is fruitful. Ultimately what tonal curve we need for each and every piece of content is unknown because we don't know what the producers and talent heard in their rooms. This is why I continue to say that the Harman curve should be the starting point and you adjust based on taste.

That certainly works. It might not be optimal though and it also involves a manual step. Auto-manual EQ?

For material with mix problems, it makes more sense to use the tone controls. The room EQ is supposed to EQ the room, not the programming. It is highly unlikely that people mixing material are going to EQ the material to correct for problems in speaker response to the level of detail that room EQ does, and even more unlikely that any user of room EQ is going to tweak the curve for every program.

IMO the whole claim that the reason flat room EQ is undesirable is because it does not accurately reflect the original intent in the mixing room is a fallacy. IMO the reason flat room EQ is undesirable is that it violates the intent of the speaker designer who designed the speaker to sound good with a sloping down power response in an enclosed space.

Besides, Harman did not test the target curves as an independent variable so your statement is only a supposition. For someone who is normally quite specific in demanding science, your implicit assumption that this experiment validated the target curve of the winning system is out of character. There were many more variables involved and the most striking one is the smoothness and extension of the Harman curve. That could have accounted for a lot of the preference. In fact their own conclusion states exactly that.

Numerical analysis to determine something that ultimately has no calibration at the source is a fruitless endeavour.

The speaker itself has excellent calibration. Pro speakers are reference audio generators. That is why I propose measuring the speaker in the room to determine the proper curve in the room.

Here, check this out:



I filtered out everything but Harman EQ and no EQ using a bitmap editor.

Look that over and tell me Harman did not just draw a straight line through the in-room response. From my perspective, they did exactly what you are telling me is the wrong approach and at least one person who claimed to know supports that supposition.

I think it is worth some investigation. If I had the resources I would just do it, patent it, and laugh (or cry, depending on if it worked) all the way to the bank.

Since that information does not exist as a practical matter for anything other than Harman speakers, Auto EQ systems still need to be performant. Hence, when we find something that works for majority of people, we should use that. And the sloping down target curve is one of those. It has been shown over and over again to be the right strategic choice.

...but it has never been compared to my proposal.

I have been making the same claim to support Audyssey in budget systems. It works well enough for the purpose and many people lacking in bottomless pockets agree with me. Does that mean it is time to rest on my laurels?

My proposal would also be a sloping down target curve when a speaker is measured in a room but the slope and curvature would be measured directly off the speaker where it sits.

Harman speakers with custom EQ is actually a great idea. Any known irregularities in the on-axis response can be EQd out by design. In that respect it has a distinct advantage over my proposal, but I still think there is also value in setting the target or at least tweaking it from the measured in-room response. For all we know maybe that is exactly what that JBL EQ for Harman speaker is already doing. Have you asked them?

I think you misunderstanding me. I am not proposing to rewrite the book on EQ based on a guess. Only an impulsive idiot does something like that. I am identifying what I perceive as unanswered questions and speculating on a new approach and proposing an experiment like Harman's only more sophisticated in its approach to the problem rather than empirical guess based on listener preference (if that is actually how it happened... really hate all this mindreading).

When I said I would blow a million on my proposed EQ it is because I like to experiment, not because I want to dictate. OK?

Anyway Onkyo seems to have already adopted something similar with AccuEQ so it seems that it is coming whether anyone wants it or not.

Harman tested EQ systems with widely varying performance (not just widely varying targets) and it did not compare straight line to B&K curved line nor did it attempt to vary the slope or shape of their straight line nor did it attempt to determine if there were any correlations between in-room non-EQd response and the best target across multiple types of speakers in multiple placements in multiple rooms.

Their justification for not checking on all of this was... mono listening is more accurate? B&W sells the most pro speakers? No, it was probably because they ran out of budget, time, and/or patience.

Or maybe just like already indicated by others on this thread they already knew about everything I am proposing to check into and either quietly slipped it in to the target curve or just ignored it deliberately because of the complexity of getting it to converge reliably. Who knows? I am no mind reader and there are proprietary technologies involved.

Do you have the actual Harman EQ they used in that study? Does it automatically adjust the slope of the target based on in-room response? Have you tested it for any changes in target due to measurement? Maybe everyone looks at that study and just guesses that the slope is a fixed target. I do not know, I am not in that league.

I am simply proposing to check into it. I am proposing to finish what Harman started instead of just leaving it status quo because it was good enough last year.

As far as people not being able to discriminate EQ well with multiple speakers, there are several likely possibilities to explain that.

One is the 'euphonic' experience of surround sound that you mentioned distracting listeners from the sound of a speaker.

Another is the filling in of gaps in the spectrum with multiple sources stimulating multiple room modes and multiple reflections etc. just like with multiple subwoofers.

Another is the application of the same EQ target curve to all speakers regardless of type and placement when they actually need different target curves to optimize them individually.

There are at least three possible explanations for why it is easier to discern frequency response with only one speaker. I think it is worth asking the questions in a scientific experiment. The answer could have profound implications for the audio market.

What I am really saying here is why didn't Harman try everything possible to optimize the target? They certainly did a great job with the implementation. That measured response is straight as an arrow. Never seen anything like it anywhere else.

Every time I see someone defending that target it is always because Audyssey flat or flattish EQ with its choppy, bandlimited measured result came last in preference. That does not validate Harman's target curve as the optimal approach to room EQ. All it does is make Harman EQ superior to Audyssey under those test conditions.

Here is a nice thought experiment. Put the same sound system into a huge anechoic chamber that absorbs down to 10Hz. EQ it with Harman's treble-attenuated response and then EQ it with Audyssey flat MultEQ-XT32.

Listen to it on-axis.

Which one wins the preference test now?

If that thought experiment does not at least raise some doubt about the wisdom of always using a fixed target curve as the first approximation and ignoring the actual in-room response, then I might as well check myself into an insane asylum because it means I have totally lost my marbles.