How important is good speaker design if you have Digital Room Correction

[Ellebob]

If your on axis response and off axis response from reflections are not similar, EQ simply can not fix this. We hear a combination of direct and reflected sound. when those are similar it adds to the sound making it more pleasant, when those are different it makes it less pleasant. You can not EQ the combined sound because you still have decorrelation between the direct and reflected sound.

You can absorb reflections if off axis response is poor. However you do need thick panels of at least 4 inches and preferably mounted with an inch or two space to get almost the full frequency range. However, when you do this you get great focus but lose envelopment which is often very pleasing to have good envelopment.

Now, if you have a speaker with an off axis response that is similar to its on axis response but doesn't have your desired frequency response this can be fixed with EQ. Sometimes I think the characteristic of similar on/off response as well as bass capability are bigger factors that influence peoples preference when deciding speakers then flatter frequency response.

 

[Sancus]

This sounds important but I cannot quite grasp it.
In order for me to understand, can you confirm if on axis and off axis response simply sum up at the listening spot to get the measured response?

That's correct. If you don't do time gating, your microphone gets the sum of all sound and it cannot distinguish.

However, your ears can distinguish between reflections and direct sound. This core notion, that there is a difference between an in-room measurement and what your ears hear is one of the primary reasons behind the spinorama measurement technique and the conclusions about the importance of off-axis response. The thread thewas linked above contains more information about this, as does this PDF written by Toole himself:

"So, why not just measure the loudspeaker in a room? Because, compared to two ears and a brain, a microphone is a “dumb” device. It accepts sounds from any angle, at any time, and treats them equally. In contrast, a human distinguishes between direct sounds and later arrivals (the precedence effect and forward masking), and between sounds from one direction and those from another (binaural discrimination)."

Of course, if you want the whole story, you should really read the book that explains it all. It is long, but, if you truly want the comprehensive explanation and don't want to just take things as fact from forum posts, it's worth getting through.

 

[o2so]

If your on axis response and off axis response from reflections are not similar, EQ simply can not fix this. We hear a combination of direct and reflected sound. when those are similar it adds to the sound making it more pleasant, when those are different it makes it less pleasant. You can not EQ the combined sound because you still have decorrelation between the direct and reflected sound.

You can absorb reflections if off axis response is poor. However you do need thick panels of at least 4 inches and preferably mounted with an inch or two space to get almost the full frequency range. However, when you do this you get great focus but lose envelopment which is often very pleasing to have good envelopment.

Now, if you have a speaker with an off axis response that is similar to its on axis response but doesn't have your desired frequency response this can be fixed with EQ. Sometimes I think the characteristic of similar on/off response as well as bass capability are bigger factors that influence peoples preference when deciding speakers then flatter frequency response.

Thank you, this is great and you are answering to my question.

Let' s see if I understand.
As the frequency response at the listening position is a sum of direct and reflected sound, it means that Dirac will have to make some sort of assumption on how much it is due to direct sound and how much to indirect, so it can calculate the correction to be applied. This is true regardless of the two FRs being similar, or different.

For instance, let's assume that there are -6db measured at the listening position at a given frequency.
If direct and reflected sound contributed 50-50 to the measured signal (for instance), these could be the result of the following combinations (for instance):

CASE1
Direct: -1Db
Indirect: -5DB

or
CASE 2
Direct: -3db
Indirect: -3db

If DIRAC made the correct assumption that direct and indirect sound contribute 50-50, it would add 3db to the output at that frequency, so that:

CASE 1 becomes:
Direct: (-1 +3)=+2Db
Indirect: (-5 + 3)=-2DB
And we measure +2-2 = 0db

CASE 2 becomes:
Direct: (-3+3) = 0db
Indirect: (-3+3) =0db
And we measure +3-3 = 0db

In both cases, we get rid of the dip at the listening position.

The only thing that can go wrong is if Dirac makes the incorrect assumption on the relative contribution of direct and indirect sound, for instance, if the assumption is 50-50 while in reality, we have that direct sound contributes for 70% and indirect for 30%. In this case we would have the following result:

CASE 1 becomes:
Direct: (-1 +3)=+2Db
Indirect: (-5 + 3)=-2DB
But we measure +2(x0.7) -2 (x0.3) = not 0db

CASE 2 becomes:
Direct: (-3+3) = 0db
Indirect: (-3+3) =0db
And we measure +3 (x0.7) -3 (x0.3) = 2.1 + 0.9 = still 0db!

So, with uneven off-axis FR we rely on Dirac making the correct assumption on how much direct and indirect sound contribute to the measured response. And this is - I take it - the source of the problem.

Am I getting this right?

 

[flipflop]

Am I getting this right?

No.
The in-room response isn't a simple arithmetical sum of the on-axis and off-axis frequency responses. It is more closely resembled by the off-axis response.
You have no idea what's happening to the direct sound if you make EQ adjustments based on the in-room response. That's why we need anechoic data.

 

[o2so]

No.
The in-room response isn't a simple arithmetical sum of the on-axis and off-axis frequency responses. It is more closely resembled by the off-axis response.
You have no idea what's happening to the direct sound if you make EQ adjustments based on the in-room response. That's why we need anechoic data.

uhm ok. Isn't this equivalent to saying that it is a sum, but one in which off-axis response has a much greater weight than on-axis? I am just trying to understand how DIRAC thinks

 

[Pio2001]

This sounds important but I cannot quite grasp it.
In order for me to understand, can you confirm if on axis and off axis response simply sum up at the listening spot to get the measured response?

It is rather this way : the measurement sums the on axis and off axis response and displays the result. But, at the listening location, on axis and off axis response are physically distinct : on axis comes first, and off axis arrives several milliseconds later.
The result is beginning of attacks sounding one way, and their decay sounding another way.

 

[Sancus]

uhm ok. Isn't this equivalent to saying that it is a sum, but one in which off-axis response has a much greater weight than on-axis? I am just trying to understand how DIRAC thinks

I don't know too much about what exact math Dirac does(it is a proprietary algorithm that is constantly being changed, for one thing), but typically with DRC the frequency response is corrected to reach the target curve AS MEASURED, and no further assumptions are made about direct/reflected sound.

Whether or not that target curve is an improvement depends on the speaker; it's far more likely to be an improvement if it's a good speaker with even off-axis response. If it's a speaker with wildly unpredictable off-axis response, then you could easily make things WORSE by trying to correct the in-room measurement.

In general, DRC is best used to correct low frequencies only, where most room issues lie. To use it to correct full-range speaker response you need anechoic data and for lower end speakers, component variation adds an additional large variable. It's not uncommon to have 2-3dB(or even more!) variation with cheaper tweeters, so trying to EQ their response eg. based on the data in Amir's reviews would not necessarily be as easy as some posts would make you believe.

If you have a good speaker to begin with though, you can apply gentle shelf filters to higher frequencies to change tonality if you wish and expect fairly good results.

 

[NTK]

uhm ok. Isn't this equivalent to saying that it is a sum, but one in which off-axis response has a much greater weight than on-axis? I am just trying to understand how DIRAC thinks

Room reflections aren't minimum phase behaviors. The direct sound and reflected sound do not simply add together. For example, if a reflection (reflected sound) has a 1 ms delay, what is added to the direct sound is the reflected off-axis sound radiated from the speaker 1 ms in the pass. And we have many many such reflections. Steady state measurements cannot reveal the complex relationships, and they cannot be (fully) corrected by EQ.
https://www.roomeqwizard.com/help/help_en-GB/html/minimumphase.html

[Edit] Illustration using a decaying sine wave: Y1 (blue) is the direct sound, Y2 (orange) is the reflected sound -- same magnitude but with a delay. Steady state measurement will simply give us the red curve = 2*Y1, which is not the same as the green curve Y1 + Y2. (Difference is the dotted purple curve.)

[Edit] I made a math error in my original graph. The magnitude from the "steady state measurement" was incorrect as I forgot to account for the phase shift. The corrected error wasn't bad at all until the reflection is very close to 180° out-of-phase with the direct sound. The "steady state measurement" would be zero if the reflection is of the same strength as the direct sound as they cancel each other. With a delay, the sum is not zero.
Y1 (blue) is the direct sound, Y2 (orange) is the reflection -- same magnitude with 180° phase lag. Steady state measurement will give the red curve (zero), green curve is the actual sum. This is the worst case scenario.

 

[Ellebob]

While measurements have the capability to determine direct response and reflected response using gated measurements as well as figuring out distances for direct and reflected sound it still can't correct it. So yes it can correct the frequency response of the combined sound so it is correct at the listening position. So while it might have the perfect frequency response at the listening position it still sounds bad. Dirac or any other EQ can not change that there is a difference between the on and off axis response even if the combined response is good. Our ears pick up the difference and don't like it. This is why you will see some people prefer the sound with EQ and others prefer to turn it off. EQ cannot fix all acoustic problems.

I consider EQ to be the icing on the cake. It will make a good system sound better but it rarely fixes a bad system. Gerry Lemay (found of Home Acoustics Alliance) which teaches calibration of audio says "Getting good sound is easy, just follow the rules." Rule number one is start with good speakers because if you don't have that everything else you do is not going to matter. You don't have to spend a lot to get good performance.

 

[o2so]

Room reflections aren't minimum phase behaviors. The direct sound and reflected sound do not simply add together. For example, if a reflection (reflected sound) has a 1 ms delay, what is added to the direct sound is the reflected off-axis sound radiated from the speaker 1 ms in the pass. And we have many many such reflections. Steady state measurements cannot reveal the complex relationships, and they cannot be (fully) corrected by EQ.
https://www.roomeqwizard.com/help/help_en-GB/html/minimumphase.html

[Edit] Illustration using a decaying sine wave: Y1 (blue) is the direct sound, Y2 (orange) is the reflected sound -- same magnitude but with a delay. Steady state measurement will simply give us the red curve = 2*Y1, which is not the same as the green curve Y1 + Y2. (Difference is the dotted purple curve.

View attachment 101264

thanks. DOes this not mean that even with speakers having a flat off-axis response, DIRAC will not be able to successfully correct EQ because it only measures the off axis that gets to the mic which was radiated 1ms in the past?

 

[Dimifoot]

Discussion might get more interesting if we consider the case of multichannel/immersive setups (loudspeakers all around you,like an Atmos setup- so mainly direct sound at the listening area) in well-treated dedicated rooms.

We should also consider that in our days is easy to find speakers with relatively smooth DI. But still dynamic range and limited distortion is expensive to buy. And you might need those for DRC.

Under these conditions the role of DRC becomes more and more important.

 

[NTK]

thanks. DOes this not mean that even with speakers having a flat off-axis response, DIRAC will not be able to successfully correct EQ because it only measures the off axis that gets to the mic which was radiated 1ms in the past?

That's the reason it is normally recommended not to EQ above the Schroeder frequency. But if we have anechoic measurements, like those provided by Amir, we can EQ the on-axis or listening window response to flat to at least get the benefits of a neutral direct sound.

 

[sfdoddsy]

Basically you need to EQ two things.

The first is speaker response, and requires close miked/gated measurements. This fixes problems inherent in your speakers natural response.

The second is room response from the listening position.

Dirac only does the latter. You can do both with a MiniDSP or even better DEQX.

But, as mentioned, if the directivity is poor the results may not be what you wish for.

Speakers like Neumann have built-in speaker DSP and good directivity so are ideal for simply using Dirac below the Schroeder frequency.

 

[dfuller]

Still very much so. The thing that bothers me, though, is that you think EQ is enough to actually correct for a room. It's not. EQ can't correct for longer or shorter RT60s than desired.

 

[o2so]

That's the reason it is normally recommended not to EQ above the Schroeder frequency. But if we have anechoic measurements, like those provided by Amir, we can EQ the on-axis or listening window response to flat to at least get the benefits of a neutral direct sound.

got it, thanks. I'd be interesting to know if Dirac can actually establish a time relationship between when each impulse is generated and when it is received so that it can figure out how much is direct and how much is reflected, and correct accordingly. I have a feeling that this is exactly how it works

 

[o2so]

Still very much so. The thing that bothers me, though, is that you think EQ is enough to actually correct for a room. It's not. EQ can't correct for longer or shorter RT60s than desired.

sorry for bothering you with my ignorance. And by the way, any consideration on decay was out of scope. I know it cannot be fixed. Never suggested you could play music in a cathedral as long as you have DIRAC

 

[richard12511]

I could not quite encapsulate the topic I wanted to discuss here in the title, because it is a bit more complex than that.

ASR provides incredibly helpful information on - among other things - speaker performance. One of the key elements considered to determine such performance is Frequency Response, with a particular emphasis given - rightfully - to the Predicetd In Room (PIR) response.

This depends on the FR on-axis, off-axis and first reflections. The latter two items, correct me if I am wrong, depend on the capability of the speaker to project a relatively flat FR off-axis, so that when this bounces off the wall or the ceiling and gets to us it does not mess up the balance of the on-axis FR.
Speakers such as Revels or Genelec are particularly good at this (and also at most other things, minus the awful design IMHO). In some cases we have seen some of these speakers having a non-perfect on-axis FR, but an almost perfect PIR because the first reflections would end up FIXING the issues seen in the on-axis response.

At the same time Digital Rom Correction systems have become increasingly popular, very effective and really not more expensive than most traditional preamplifiers (I am thinking MiniDSP SHD). These systems simply measure the frequency response at the listening position and apply equalisation and time delays to ensure that what gets to your ears is a flat response, or whichever response you ask the software to generate. Not only this, but software like DIRAC fix time alignment of multiple drivers and phase.

Now, in a system that has this type of DSP, how important is the speaker natural PIR and step response, really?
I think these speaker attributes become important only if:

• one listens to music in different parts of the room, not always in the same spot. In this case, DRC can not do much as FR and step response can really only be properly fixed at one listening spot. BUT seriously, who on earth would spend thousands on a hi-fi system and then not sit right in the sweet spot to enjoy it? Yes ok we all like background music when we do other stuff around the house, but being background music its fidelity is also relatively unimportant.
• one believes that DSP is evil and will pollute your analog music with evil digital artifacts.

In addition to the above, if most of our listening is in the sweet spot and our speakers are good on-axis but not so good off-axis, we can simply treat the sidewalls, use a nice thick rug and absorb first reflections so that the DRC does not have to work too hard (not that I have ever heard my MiniDSP complaining though).

Finally, even if we have speakers with near-perfect PIR that do not require DRC (this is really a non-possible case, because even these speakers will massively benefit from DRC for the lower frequencies), is there not a time-smear issue due to the high frequencies being "sprayed" (like Mr Sanders would say) all over the room and bouncing back at us at different times? Is it not better to absorb first reflections ANYWAY?

So, if one:

• uses DRC on the full frequency range
• only or mostly sits in the sweet spot
• has some floor and side-walls absorption

How important is to have speakers with great diffusion and PIR?

It's a good question. I think that if you're good with EQ, frequency response becomes significantly less important, but directivity still matters. Take a speaker like the KEF LS50. It has some real frequency response errors in some very critical ranges, but it has excellent dispersion characteristics and is a point source design. I would bet a speaker like that sounds very near SOTA (at least at moderate volumes) after you EQ out the frequency response errors, but this does require some technical proficiency from the end user. I'm sure there are many speakers that sound better than the LS50 out of the box, but probably relatively few that can best it post EQ(again, at moderate volumes).

Speaking of the LS50 and EQ. You may be interested in this excellent article from member @mitchco . It does a great job of showing how far EQ can take you in terms of equalizing speakers when you're starting with two very good speakers with different(but great) dispersion characteristics.

 

[NTK]

got it, thanks. I'd be interesting to know if Dirac can actually establish a time relationship between when each impulse is generated and when it is received so that it can figure out how much is direct and how much is reflected, and correct accordingly. I have a feeling that this is exactly how it works

I don't know for sure but I suspect not. I think it is a bit difficult for the software to automatically figure out when the reflected sound starts arriving.

 

[Pio2001]

[Edit] Illustration using a decaying sine wave: Y1 (blue) is the direct sound, Y2 (orange) is the reflected sound -- same magnitude but with a delay. Steady state measurement will simply give us the red curve = 2*Y1, which is not the same as the green curve Y1 + Y2. (Difference is the dotted purple curve.)

View attachment 101264

Steady state measurement (using pink noise, for example) can give you none of the above, as it doesn't record time-domain information.

It will however give you the same frequency response as the one of the green curve.

 

[o2so]

I don't know for sure but I suspect not. I think it is a bit difficult for the software to automatically figure out when the reflected sound starts arriving.

is it? uhmmm There would be frequency x emitted at time 0. Then frequency x received at time 1, then again at time 2, then again at time 3. However, I suspect these would largely overlap to the point that it is not possible to say which is what