The Audyssey MultEQ Editor app users thread (with facts and tips)

[PENG]

The nice thing about this is that Audyssey is still in charge of creating the filters. I believe by manually entering the control points simply help Audyssey to do a better job of creating more/accurate filters than the auto process that has to rely on the operator, and limited by the resolutions even with the 4X improvements over XT in the bass band. Audyssey auto doesn't have the benefits of seeing the actual effect shown by REW/Mic. So we are just providing a feedback loop in effect by entering the new control points based on what we see on the REW graphs. So you can't mess it up if you follow the REW graphs hints and enter the control points in a logical fashion.

The most important thing to do is to avoid entering too much boost, as Jon AA allude to. The key is, try improving things with cuts, and only boost if there is no other way. I learnt that from using the mini/REW equalizers too, you could make things much worse, and/or unstable if you boost too much, anything more than 3 dB should be avoided as much as possible, you never know if Audyssey has already applied the maximum boost at the same points, until you see some totally erratic results.

 

[Matthew J Poes]

I spent a full day yesterday to play with the Editor App and REW, the best curves obtained within a bubble of 8 to 12 inches at the MMP, left, right, above and below were surprisingly good. I know it is academic, whether the FR from 15 to 200 Hz is within +/- 1.5 dB or +/- 3.5 dB with 1/12 smoothing is not audibly different to me. It just shows the App does work well if you take the time, and use the Ratbuddyssey UI.

@Pogre , having seen your graphs, I am quite sure you can do the same or better than me without the miniDSP. If you can spend one to two days on it, you will end up having the mini for sale and spend the money on stead and beer. Will post some screen shots of the target curves later, time permitting.

@Jon AA , @DJ7675, would love to see some graphs from you guy experts.

@Matthew J Poes , I viewed your Y-tube video a couple times and am interested to know exactly how you did your listening window plots, did you actually measured the angles, or just eyeballed and estimated the angles for each?

I am curious because I have moved the mic up to 12 to 18 inches in any direction, average the plots and I could not see Audyssey messed anything up. I know you have expensive mics and stuff but you did use REW right. Just can't understand how the results could be so different. And, have you tried using the App to squeeze the last 1-3 dB improvement?


View attachment 36251

View attachment 36252

The answer is because you measured in a room to see if it messed up. You can't see in the room what its doing to the listening window.

So first, how I did it, I took a speaker we have previously measured and used the raw data to calculate its predicted in-room response, listening window, early reflection, and everything else from Spin-O-Rama. Then I took that speaker, ran Audyssey, captured the transfer function of the Audyssey correction, and applied it to the speaker in room and to the raw original data. That is how I showed you what it did.

Now Gene was concerned that that and a model I ran wouldn't be believable. So I did take the receiver outside with the speaker to confirm, and I still had a very close match in the listening window.

Now, you might say, but if I can't see it in room, then how does this outdoor thing matter? That was the point of the video and requires that you understand how the ears and brain work vs a mic. Hence my first article I ever wrote for Audioholics. A lot of people assume that a mic and our ears pickup and process sound in the same way. That is simply not true and that is why both Earl Geddes and Floyd Toole talk about this quite a bit in their respective books. Toole uses his catch phrase, Two Ears and a brain quite a bit when describing the difference.

An omni mic cannot tell from what direction sound is coming from, so even with advanced FFT analysis and windowing, it is impossible for it to ignore reflections the way a brain and ears do. If the mic picks up sound from multiple directions but with a similar time signature, it combines the two and cannot seperate or process them out. A 3D mic and special software can actually discern and process out these reflections, but an omni mic like you are using with REW and REW itself (a single channel FFT) cannot.

How does the brain and ears work differnetly? The ears are directional with a forward facing directional lobe. They are stereo as well, we have two. By relying on the phase/timing differences between ears, the amplitude differences, the HRTF, we can discern from what direction sound is coming from. We have evolved to be able to use this to pay attention to direct sound over reflected sound, meaning our brain actually ignores the reflections in the room that are of a similar time signature to the direct sound (obviously delayed, but it falls within what is called an integration zone where, typically, the brain might actually integrate them all like the mic does). If these early reflections are coming from the exact direction of the speaker, we can't ignore them. But they aren't, they are typically coming from above, below, to the side of, and behind us. We can largely ignore many of those.

So what we know is that for bass frequencies below the Schroeder frequency, we can't discern reflections and direct sound in a room. The wavelength of bass frequencies relative to the temporal differences of the bass reflections are too small for the brain to separate, so it doesn't. That means what we measure with a microphone and what we hear is largely the same thing below ~100hz.

A transition zone exists between this largely modal region and the stochastic region and that is not a hard and fast zone, but let's call it between 100hz and 1khz, but mostly below 500hz. In this region, we transition slowly from a complete inability to discern reflections to ever greater ability to discern reflections. In this region, some of what you see in a microphone is what you hear, but some of it is not.

Above this point, largely that point above 500hz, the microphone is combining direct and reflected sound in a manner that does not accurately track what you hear.

EQ systems cannot fix this. Because they rely on omni mics they can't really tell what direction the reflections are coming from and cannot know what you can and cannot hear. So the algo's either ignore this and make big mistakes or they are setup to make corrections that are more consistent with what we hear. For example, they might use a kind of windowing and some boundaries around how EQ is applied to better focus on speaker resonances vs room resonances above a certain frequency.

Audyssey has been known for some time to fall into the former camp, it just ignores this. Thus is makes more errors with speakers that have an inconsistent off-axis response. Dirac incorporates more of the latter, but I've found it to be a mixed bag as well. Depending on how bad the speaker is, it too makes mistakes.

This is an inherent problem with all room based EQ, we shouldn't be eqing a speaker flat based on in-room measurements above the transition zone. We really should be eqing the speakers natural response flat/smooth, and that is it.

A secondary problem that arises is that the "curve" is actually a natural shape that happens when a perfectly flat speaker with a response from 5hz to 20khz is then put into a normal room with average reflectiveness. What that means is that depending on the speakers actual response, its directivity, and the rooms acoustics, a different room curve is desirable. EQ systems have no way to know this and neither would consumers. You would have to a) know your rooms acoustics, b) know your speakers predicted in room response from full Spin data, and c) know how to read in-room measurement data to create a proper room curve.

Having just gone through this with Audyssey, ARC, and Dirac I can say this, all three failed with the speakers in use. One system had Revel and the other had ML. Audyssey and ARC were especially problematic with ML and required that I manually adjust it dramatically to get a correct room curve (which the owner much preferred).

 

[Matthew J Poes]

The nice thing about this is that Audyssey is still in charge of creating the filters. I believe by manually entering the control points simply help Audyssey to do a better job of creating more/accurate filters than the auto process that has to rely on the operator, and limited by the resolutions even with the 4X improvements over XT in the bass band. Audyssey auto doesn't have the benefits of seeing the actual effect shown by REW/Mic. So we are just providing a feedback loop in effect by entering the new control points based on what we see on the REW graphs. So you can't mess it up if you follow the REW graphs hints and enter the control points in a logical fashion.

The most important thing to do is to avoid entering too much boost, as Jon AA allude to. The key is, try improving things with cuts, and only boost if there is no other way. I learnt that from using the mini/REW equalizers too, you could make things much worse, and/or unstable if you boost too much, anything more than 3 dB should be avoided as much as possible, you never know if Audyssey has already applied the maximum boost at the same points, until you see some totally erratic results.

I actually want to correct two things you say here Peng.

First, Audyssey and all EQ's don't need this "feedback" loop. That isn't how it works. As I showed in my presentation, you can nearly perfectly predict the effect of EQ on a system by knowing its minimum phase system and the excess phase. With that, a minimum phase transfer function can be applied and what you get out is know. The fact that you are taking measurements, tweaking, and seeing something "better" is just you making adjustments based on your brains own internal Algo. Audyssey already knows exactly what comes out and made those decisions intentionally. Your changing what decisions it can make based on what you prefer to see. In theory, you could change Audyssey to do that initially.

As for boost, boost is a problem, big one. I know you already know this, but I want to add that I've seen Audyssey and Dirac burn out tweeters and cause woofers to exceed xmax trying to fit their response to an unatural curve.

In one case, a user was using Audyssey with Kef Reference in-wall speakers that were behind an AT screen. The AT screen was attenuating the treble and Audyssey attempted to correct this. It burned out the tweeters over and over again. When I helped this user, we changed the target curve and as far as I know he has never had a problem since.

A JBL speaker that was under review for another magazine was being corrected with Dirac. The standard curve that it fitted attempted to flatten out and extend the bass and in effect caused the woofers to easily exceed xmax. It caused horrible audible distortion. Adding more spline points to the curve and rolling the bass off solved that problem.

I've also done experiments where I looked to see what these EQ's do when they apply a significant correction and then run through a dummy load. I did a ON vs Off comparison of the amplifiers clipping and found the amps were clipping a lot more easily at real world levels. The test cases were intentionally extreme, where I made it flatten and extend the bandwidth, but it did show how severe the problem is.

I'm not anti-EQ, just against its misuse. My goal has been to educate people into not falling into a false sense of ability around what EQ can do. It's far less magical than people think.

 

[PENG]

I actually want to correct two things you say here Peng.

First, Audyssey and all EQ's don't need this "feedback" loop. That isn't how it works. As I showed in my presentation, you can nearly perfectly predict the effect of EQ on a system by knowing its minimum phase system and the excess phase. With that, a minimum phase transfer function can be applied and what you get out is know. The fact that you are taking measurements, tweaking, and seeing something "better" is just you making adjustments based on your brains own internal Algo. Audyssey already knows exactly what comes out and made those decisions intentionally. Your changing what decisions it can make based on what you prefer to see. In theory, you could change Audyssey to do that initially.

As for boost, boost is a problem, big one. I know you already know this, but I want to add that I've seen Audyssey and Dirac burn out tweeters and cause woofers to exceed xmax trying to fit their response to an unatural curve.

In one case, a user was using Audyssey with Kef Reference in-wall speakers that were behind an AT screen. The AT screen was attenuating the treble and Audyssey attempted to correct this. It burned out the tweeters over and over again. When I helped this user, we changed the target curve and as far as I know he has never had a problem since.

A JBL speaker that was under review for another magazine was being corrected with Dirac. The standard curve that it fitted attempted to flatten out and extend the bass and in effect caused the woofers to easily exceed xmax. It caused horrible audible distortion. Adding more spline points to the curve and rolling the bass off solved that problem.

I've also done experiments where I looked to see what these EQ's do when they apply a significant correction and then run through a dummy load. I did a ON vs Off comparison of the amplifiers clipping and found the amps were clipping a lot more easily at real world levels. The test cases were intentionally extreme, where I made it flatten and extend the bandwidth, but it did show how severe the problem is.

I'm not anti-EQ, just against its misuse. My goal has been to educate people into not falling into a false sense of ability around what EQ can do. It's far less magical than people think.

Thanks for sharing, there are lots to digest, will save it for weekend reading.. Suffice to say I was only kidding about the "feedback loop" thing, but I guess you couldn't tell obviously as I did a good job. Very good job on that video, thinks again.

 

[Pogre]

I actually want to correct two things you say here Peng.

First, Audyssey and all EQ's don't need this "feedback" loop. That isn't how it works. As I showed in my presentation, you can nearly perfectly predict the effect of EQ on a system by knowing its minimum phase system and the excess phase. With that, a minimum phase transfer function can be applied and what you get out is know. The fact that you are taking measurements, tweaking, and seeing something "better" is just you making adjustments based on your brains own internal Algo. Audyssey already knows exactly what comes out and made those decisions intentionally. Your changing what decisions it can make based on what you prefer to see. In theory, you could change Audyssey to do that initially.

As for boost, boost is a problem, big one. I know you already know this, but I want to add that I've seen Audyssey and Dirac burn out tweeters and cause woofers to exceed xmax trying to fit their response to an unatural curve.

In one case, a user was using Audyssey with Kef Reference in-wall speakers that were behind an AT screen. The AT screen was attenuating the treble and Audyssey attempted to correct this. It burned out the tweeters over and over again. When I helped this user, we changed the target curve and as far as I know he has never had a problem since.

A JBL speaker that was under review for another magazine was being corrected with Dirac. The standard curve that it fitted attempted to flatten out and extend the bass and in effect caused the woofers to easily exceed xmax. It caused horrible audible distortion. Adding more spline points to the curve and rolling the bass off solved that problem.

I've also done experiments where I looked to see what these EQ's do when they apply a significant correction and then run through a dummy load. I did a ON vs Off comparison of the amplifiers clipping and found the amps were clipping a lot more easily at real world levels. The test cases were intentionally extreme, where I made it flatten and extend the bandwidth, but it did show how severe the problem is.

I'm not anti-EQ, just against its misuse. My goal has been to educate people into not falling into a false sense of ability around what EQ can do. It's far less magical than people think.

So what about limiting what Audyssey does to say, 200 hz and down with the app editor? That's my focus with using EQ. I think Audyssey does a great job with my subs. I know it sounds a lot better than direct to me.

 

[Jon AA]

@Jon AA , @DJ7675, would love to see some graphs from you guy experts.

Well, I certainly wouldn't call myself an Audyssey "expert." More of an Audyssey Ninja!

Most of my measuring lately has been experimenting/learning/researching...but here are a couple. They're obsolete already but that's OK....

My left front set to large:

With Sub:

Breakdown of the integration (60 Hz crossover):

Those are all steady-state room responses at the MLP, 5 measurement small spacial averages (about a 1 foot box around my head), smoothed 1/12 octave.

 

[Jon AA]

The most important thing to do is to avoid entering too much boost, as Jon AA allude to. The key is, try improving things with cuts, and only boost if there is no other way.

A couple of notes on this. Yes, generally correct. However, you need to be cognizant of "Boost--relative to what?"

Remember, the "boost" you put into ratbuddy is relative to the Aud Ref curve--not your speakers' natural response. For example, say you have a large set of speakers set up relatively close to the front wall--or even in a corner. Due to boundary/room gain, the natural response of your speaker may have a huge amount of bass. The Audyssey reference curve will cut that flat--it may be cutting more than 10 db in some situations.

In that case, even if you add "boost" of 4-6 db on the bottom end to the reference curve as quite is common, that ends up in actuality being a "cut." One needs to compare what you're doing with measurements with Audyssey switched off to keep you out of trouble.

And on the high end--if you have speakers that are flat or bright on the very top end, Audyssey will apply a rolloff on the top end, cutting quite a bit. So even decent sized "boosts" in that area may still be cuts to the natural response of your speaker.

Edit: One more note about bass and boost--if you will ever run your speakers set on "large" and play them loud, if they are ported it's a really good idea to determine their port tuning frequency. Do this by putting the mic about an inch away from the woofer, exactly in the center. It'll be easy to see where it dips:

For that speaker, you can see it's about 39 Hz. Generally, Audyssey does an OK job of recognizing the -3db downpoint of your speaker and not applying boost below that. However, in a case like I described above (speaker stuck in a corner, etc), your in-room response could be flat to the mid 20 Hz range. In that case, Audyssey would allow boosting below the port tuning frequency. That's a big no-no.

So if you will ever run the speakers full range really loud, I'd advise drawing yourself in a high pass filter at 40 Hz in Ratbuddy to protect your speakers. If you're always going to run them as small crossed to a sub above 40 Hz, it's not something you need to worry about as a small amount of boost at frequencies above that won't be the end of the world.

 

[PENG]

Well, I certainly wouldn't call myself an Audyssey "expert." More of an Audyssey Ninja!

Most of my measuring lately has been experimenting/learning/researching...but here are a couple. They're obsolete already but that's OK....

My left front set to large:

View attachment 36357

With Sub:

View attachment 36358

Breakdown of the integration (60 Hz crossover):

View attachment 36359

Those are all steady-state room responses at the MLP, 5 measurement small spacial averages (about a 1 foot box around my head), smoothed 1/12 octave.

Nice! I did a 10" left, center, right, above and below listening bubble today (8" only for below MMP), so 9 positions.

 

[Jon AA]

Thanks for all that, Matthew. I agree 100% with everything you said. Unfortunately people are becoming even more difficult to convince of this--mainly because "Room EQ is Fun!" They don't want to believe it.

This is an inherent problem with all room based EQ, we shouldn't be eqing a speaker flat based on in-room measurements above the transition zone. We really should be eqing the speakers natural response flat/smooth, and that is it.

And wouldn't it be nice if we could do that? People often mis-state that Floyd Toole is "against EQ" or against it at high frequencies. That's not true:

What he's against, is trying to do that with steady-state in-room responses. It just can't be done correctly as you have noted.

Kevin Voecks said that pretty much any passive speaker, including their top of the line Revels, can be improved with high resolution EQ at high frequencies--but based upon anechoic data, not in-room response. That's the basis for their "Anechoic EQ" in the SDP-75. It uses anechoic data to "fix the speaker" with high resolution EQ, and then the "room EQ" portion is only used for low resolution, smooth tonal balance changes to the user's preference.

I'm working on a method for advanced users to use Audyssey XT32 as a "poor man's" (very poor man's) version of that which you may or may not approve of. It's such a shame to have thousands of FIR filters at your fingertips and leave them unused by limiting correction to the transition frequency when they could actually improve your speakers if you just told them what to do. And some speakers need a lot of improving! But even good speakers generally leave quite a bit of room for improvement. Telling all those filters what to do, that's the tricky part....

Most people won't have good anechoic data of their speakers. I'm putting forth that people can get good enough quasi-anechoic data with REW or Omnimic to improve their speakers at high frequencies with the right measuring techniques. Most people aren't going to be able to drag their speakers (and AVR) outside to do proper measurements free of reflections and that will limit how low they can get useful data (the lower frequencies require more reflection-free space). But since at very low frequencies you want to EQ the speaker and the room together (the way Audyssey does it) anyway, it sort of all works out (with a slight grey area in the middle).

By measuring several speakers from various locations in the room, I've convinced myself that if you use gated measurements of around 4 ms at 1 meter, one can get consistent, repeatable results down to around 800 Hz or so without room reflections interfering. Here's a measurement of a speaker in the center of my room and one with it pushed up against a wall:

(That's using omnimic in "blended" mode set at 4 ms, no smoothing at all.)

As you can see, above about 800 Hz, the results are basically identical (I was actually surprised the first couple times I did it that in that I could set the mic up that consistently in different locations multiple times). I think with careful technique, some may get useful data even a little below that--depending upon their speakers, how big their room is, etc. With these particular speakers (large MF and HF horns with a decent distance between the drivers) I don't really trust the data much below that at 1m (I don't even know if I can believe that dip at 950 Hz, I think it may disappear with a more far-field measurement).

So, in my opinion there's still an "area of uncertainty" from about 500-1000 Hz where speaker type and measuring technique can affect the validity of the results. I'm still trying to figure out best practices for this area.

But if we forget about that zone for a minute and just focus on the higher frequencies where the measurements should be reliable, we come to the point of all this. If one uses my (rather tedious but doable) technique to fool Audyssey into doing what you want, the capability of XT32 to fix things at high frequencies is astounding, as I was alluding to above.

Here's a 1m gated measurement of the speaker "fixed" above 1000 Hz or so, with no smoothing:

As you can see, it's +/- 1/2 db from 1050 Hz to 18K+. That's a very JBL M2-like high frequency response. While XT32 generally doesn't do high resolution changes at high frequencies (as well it shouldn't--as you, Toole, ect have pointed out it doesn't have enough information from steady state room measurements to do it properly) it certainly has the capability to do so if you tell it to. This is what I found most astounding, I had no idea XT32 would actually be able to make such fine corrections. That's a boatload of DSP horsepower so many people already own--11 channels of it in a box they already have integrated into their system.

That EQ results in the in-room measurements I posted above. Basically "room-EQing" at low frequencies, the above quasi-anechoic EQ at high frequencies and letting the "room curve" take the shape it naturally takes.

Of course many won't be left with such a nice, flat, smoothly declining room curve even if they get the high frequencies perfectly flat at 1m--I happen to have pretty good speakers with excellent directivity control. If my speakers, for example, had a directivity mismatch at the crossover as is so common, I'd expect to see a dip in the room response there--and I'd leave it there. Many speakers have other directivity issues as well that will cause uneven in-room measurements and as you and others have noted, that can't be fixed with EQ.

But I think in most cases if one gets the frequency response smoothed and flattened quasi-anechoically, the speaker will be improved and sound as good as its basic design is going to allow. Most of the time this will be better than flattening/smoothing the room curve and screwing up the direct sound in the process.

Anyway, that's the path I've been headed down in an effort to put all that Audyssey DSP power to good use--the right way. I'm planning to write up a procedure but I'm still experimenting around to try and figure out best practices to give the average guy who might not live and breath this stuff the best results. But I'd be happy to describe the process if anybody else wants to experiment with it.

 

[DJ7675]

Thanks for all that, Matthew. I agree 100% with everything you said. Unfortunately people are becoming even more difficult to convince of this--mainly because "Room EQ is Fun!" They don't want to believe it.

And wouldn't it be nice if we could do that? People often mis-state that Floyd Toole is "against EQ" or against it at high frequencies. That's not true:

View attachment 36367

What he's against, is trying to do that with steady-state in-room responses. It just can't be done correctly as you have noted.

Kevin Voecks said that pretty much any passive speaker, including their top of the line Revels, can be improved with high resolution EQ at high frequencies--but based upon anechoic data, not in-room response. That's the basis for their "Anechoic EQ" in the SDP-75. It uses anechoic data to "fix the speaker" with high resolution EQ, and then the "room EQ" portion is only used for low resolution, smooth tonal balance changes to the user's preference.

I'm working on a method for advanced users to use Audyssey XT32 as a "poor man's" (very poor man's) version of that which you may or may not approve of. It's such a shame to have thousands of FIR filters at your fingertips and leave them unused by limiting correction to the transition frequency when they could actually improve your speakers if you just told them what to do. And some speakers need a lot of improving! But even good speakers generally leave quite a bit of room for improvement. Telling all those filters what to do, that's the tricky part....

Most people won't have good anechoic data of their speakers. I'm putting forth that people can get good enough quasi-anechoic data with REW or Omnimic to improve their speakers at high frequencies with the right measuring techniques. Most people aren't going to be able to drag their speakers (and AVR) outside to do proper measurements free of reflections and that will limit how low they can get useful data (the lower frequencies require more reflection-free space). But since at very low frequencies you want to EQ the speaker and the room together (the way Audyssey does it) anyway, it sort of all works out (with a slight grey area in the middle).

By measuring several speakers from various locations in the room, I've convinced myself that if you use gated measurements of around 4 ms at 1 meter, one can get consistent, repeatable results down to around 800 Hz or so without room reflections interfering. Here's a measurement of a speaker in the center of my room and one with it pushed up against a wall:

View attachment 36366

(That's using omnimic in "blended" mode set at 4 ms, no smoothing at all.)

As you can see, above about 800 Hz, the results are basically identical (I was actually surprised the first couple times I did it that in that I could set the mic up that consistently in different locations multiple times). I think with careful technique, some may get useful data even a little below that--depending upon their speakers, how big their room is, etc. With these particular speakers (large MF and HF horns with a decent distance between the drivers) I don't really trust the data much below that at 1m (I don't even know if I can believe that dip at 950 Hz, I think it may disappear with a more far-field measurement).

So, in my opinion there's still an "area of uncertainty" from about 500-1000 Hz where speaker type and measuring technique can affect the validity of the results. I'm still trying to figure out best practices for this area.

But if we forget about that zone for a minute and just focus on the higher frequencies where the measurements should be reliable, we come to the point of all this. If one uses my (rather tedious but doable) technique to fool Audyssey into doing what you want, the capability of XT32 to fix things at high frequencies is astounding, as I was alluding to above.

Here's a 1m gated measurement of the speaker "fixed" above 1000 Hz or so, with no smoothing:

View attachment 36368

As you can see, it's +/- 1/2 db from 1050 Hz to 18K+. That's a very JBL M2-like high frequency response. While XT32 generally doesn't do high resolution changes at high frequencies (as well it shouldn't--as you, Toole, ect have pointed out it doesn't have enough information from steady state room measurements to do it properly) it certainly has the capability to do so if you tell it to. This is what I found most astounding, I had no idea XT32 would actually be able to make such fine corrections. That's a boatload of DSP horsepower so many people already own--11 channels of it in a box they already have integrated into their system.

That EQ results in the in-room measurements I posted above. Basically "room-EQing" at low frequencies, the above quasi-anechoic EQ at high frequencies and letting the "room curve" take the shape it naturally takes.

Of course many won't be left with such a nice, flat, smoothly declining room curve even if they get the high frequencies perfectly flat at 1m--I happen to have pretty good speakers with excellent directivity control. If my speakers, for example, had a directivity mismatch at the crossover as is so common, I'd expect to see a dip in the room response there--and I'd leave it there. Many speakers have other directivity issues as well that will cause uneven in-room measurements and as you and others have noted, that can't be fixed with EQ.

But I think in most cases if one gets the frequency response smoothed and flattened quasi-anechoically, the speaker will be improved and sound as good as its basic design is going to allow. Most of the time this will be better than flattening/smoothing the room curve and screwing up the direct sound in the process.

Anyway, that's the path I've been headed down in an effort to put all that Audyssey DSP power to good use--the right way. I'm planning to write up a procedure but I'm still experimenting around to try and figure out best practices to give the average guy who might not live and breath this stuff the best results. But I'd be happy to describe the process if anybody else wants to experiment with it.

Appreciate your post and look forward to your write up on a process to use Audyssey in the manner you describe. I’ve got 3 Revel M16 and between Harman’s measurements and Amir’s measurements at ASR it would be a fun project. Just so I understand correctly, I would use Audyssey to fine tune the speaker with Audyssey (using ratbuddy). The M16 has good directivity so it would be a good example of a speaker that could be “corrected”, correct?

 

[PENG]

A couple of notes on this. Yes, generally correct. However, you need to be cognizant of "Boost--relative to what?"

Remember, the "boost" you put into ratbuddy is relative to the Aud Ref curve--not your speakers' natural response. For example, say you have a large set of speakers set up relatively close to the front wall--or even in a corner. Due to boundary/room gain, the natural response of your speaker may have a huge amount of bass. The Audyssey reference curve will cut that flat--it may be cutting more than 10 db in some situations.

You are confirming what I thought was the case. With the Editor App, and Ratbuddyssey, we can see much better the boosts and cuts at different frequencies applied in Audyssey's attempt to "flatten the curve" according to the target curve (flat in my case for the subwoofers).

In that case, even if you add "boost" of 4-6 db on the bottom end to the reference curve as quite is common, that ends up in actuality being a "cut." One needs to compare what you're doing with measurements with Audyssey switched off to keep you out of trouble.

Edit: One more note about bass and boost--if you will ever run your speakers set on "large" and play them loud, if they are ported it's a really good idea to determine their port tuning frequency. Do this by putting the mic about an inch away from the woofer, exactly in the center. It'll be easy to see where it dips:

I did some of that when building my BMR, in order to pick the port tube length. I probably should do that to find out what it is for my front mains, the Veritas 2.3i that are ported. I never run them large, and never lower than XO 80 Hz because I found Audyssey could not integrate them with my two front subs well if I set XO lower than 60 Hz. But I may still measure it just for fun. I know my little BMR's ports were tuned quite low, about 28 Hz iirc.

 

[PENG]

It's such a shame to have thousands of FIR filters at your fingertips and leave them unused by limiting correction to the transition frequency when they could actually improve your speakers if you just told them what to do.

Agreed, that's my point (made jokingly) about "feed back loop", that Matthew took it literally, when I was basically trying to say what you are saying. I meant, if we use REW/Mic to tell us the actual FR, say still restrict it to 20-200 Hz, and compare it to a flat FR, the differences at each point would represent the "error" (again at each freq point), and we can tell Audyssey what they are, that is what I meant by "feed back", and then if we (and I have done so..) use the actual data to calculate the boosts and cuts required to reach a "flat" target curve then we would be effectively, used your words, "...told them what to do...". Now, Audyssey would still not be able to achieve the targets as their software and their users are not perfect, among other factors, but I found that by repeating the same iteration process, I could eventually get the FR in the low range to about +/- 2 dB peak to peak variations/errors. In my case it has always been a very time consuming and frustration process and so I suggested that if @Pogre is interested to do it, he should set aside one full day though of course he may have an easier room and can achieve 2 dB peak to peak in less than an hour.

Also, it took me a lot of time to get one done, because I was just eyeballing things and making random decisions on the data point, almost a pure trial and error approach. I am now trying to use a spreadsheet to pick and calculate the required data points based on different target levels, and hopefully that could result in a better looking target curve reduced time to achieve the same results.

Of course many won't be left with such a nice, flat, smoothly declining room curve even if they get the high frequencies perfectly flat at 1m--I happen to have pretty good speakers with excellent directivity control. If my speakers, for example, had a directivity mismatch at the crossover as is so common, I'd expect to see a dip in the room response there--and I'd leave it there. Many speakers have other directivity issues as well that will cause uneven in-room measurements and as you and others have noted, that can't be fixed with EQ.

For me, it is just fun to figure out if I can get a flat in-room response with my speakers and subwoofers. I am not going to worry about how it sounds to me after and then re-shape the target curve accordingly. The good thing with the App is that now I can store many target curves, and I documented all the details on each curve with REW. I can listen to a different curve every day for a month, or longer if I want.

I'm planning to write up a procedure but I'm still experimenting around to try and figure out best practices to give the average guy who might not live and breath this stuff the best results. But I'd be happy to describe the process if anybody else wants to experiment with it.

Thank you for that, and please keep us posted.

 

[PENG]

The answer is because you measured in a room to see if it messed up. You can't see in the room what its doing to the listening window.

So first, how I did it, I took a speaker we have previously measured and used the raw data to calculate its predicted in-room response, listening window, early reflection, and everything else from Spin-O-Rama. Then I took that speaker, ran Audyssey, captured the transfer function of the Audyssey correction, and applied it to the speaker in room and to the raw original data. That is how I showed you what it did.

Now Gene was concerned that that and a model I ran wouldn't be believable. So I did take the receiver outside with the speaker to confirm, and I still had a very close match in the listening window.

I know you did explain it in the video, but I don't totally understand why moving the mic around in my room would not reveal any such "messed up" effects either, I thought it would, though the results wouldn't be the same as the way you did it, and I don't feel so bad now, knowing that even Gene had questions about it. I will do more reading and think harder of what you are trying to say/explain, and hopefully understand the point eventually. For now, I am seeing that no matter where I placed my omni-directional mic with a left, right, above and below by10 to 12 inches between each positions, that is, up to 24 inches between the left most/right, just as an example, the Reference FR always look much flatter than the "Off" FR curves, so I find it hard to say Audyssey would create something bad or erratic outside of the MMP.

That means what we measure with a microphone and what we hear is largely the same thing below ~100hz.

When I did my 9 position listening bubble, or window (I know you are telling me that the way I did it really didn't represent a "listening window"). I limited the range to 200 Hz as you can see the ref vs off FR curves in post#48. So now that you mentioned below 100 Hz, are you saying that at least for below 100 Hz, the way I did it would still be fine?

EQ systems cannot fix this. Because they rely on omni mics they can't really tell what direction the reflections are coming from and cannot know what you can and cannot hear. So the algo's either ignore this and make big mistakes or they are setup to make corrections that are more consistent with what we hear. For example, they might use a kind of windowing and some boundaries around how EQ is applied to better focus on speaker resonances vs room resonances above a certain frequency.

Audyssey has been known for some time to fall into the former camp, it just ignores this. Thus is makes more errors with speakers that have an inconsistent off-axis response. Dirac incorporates more of the latter, but I've found it to be a mixed bag as well. Depending on how bad the speaker is, it too makes mistakes.

I understood your (and of course Floyd's) point, but as Jon AA sort of alluded to, I also believe Floyd's (and may be yours too) point might have been occasionally and slightly taken out of context, in the sense that Floyd, you guys may not really be saying that it EQ above the transition frequency would not work, but that it could not be made to work, because the input data collected by the mic in room wouldn't be reliable at all above the transition frequency. I know this sounds confusing, but I tried..

Now, back to what I said jokingly about "feed back loop", it was actually meant to be a bit like what Jon said, but he expressed more or less the same in a clearer way. That is, if we "feed" the actual measured results back to Audyssey, by way of manually entering additional data entry points based on the actual measured in room response to alter the target curve, it would be like feeding back the error signal to the input of the control process (i.e. Audyssey auto room EQ). And I was only referring to doing it up to 200 Hz. I know Jon AA is working on making it work for the higher frequency range, I would be interested in trying to do that too eventually, but for now I just want to do the best I can, for the 20-200 hz range only.

May be I can express myself better with a numerical example. Let's say Audyssey targeted a flat response of 70 dB at 80 Hz, and the measured response was 3 dB above (i.e. 73 dB at 80 Hz as measured) the targeted level, then the error would be +3 dB, so if I enter a cut of 3 dB at 80 Hz, then I am telling, or fooling Audyssey that I want to customize my curve to have a 3 dB dip. If Audyssey is not perfect but is reasonably good, it would try to apply a cut of approximately 3 dB, and then I would get my "flat" response at 80 Hz, though Audyssey would think that I wanted a 3 dB dip instead of "flat". Now if Audyssey cannot achieve close enough to the 70 dB target, but the response did drop to 71.5 dB, then I would make a second attempt by entering a cut of another 1.5 dB; and by repeating this process many times, it would be like adding a "feedback loop" to the Audyssey EQ process, though as you said, and I knew that too, that it is not what Audyssey was designed to work and it would be in their wildest dream that some crazy user like me would so such a thing.

Having just gone through this with Audyssey, ARC, and Dirac I can say this, all three failed with the speakers in use. One system had Revel and the other had ML. Audyssey and ARC were especially problematic with ML and required that I manually adjust it dramatically to get a correct room curve (which the owner much preferred).

Just want to be clear, are you saying all three failed in the below 200, or even 100 Hz, or just the higher frequencies?

 

[PENG]

I am happy to report that the spreadsheet calculated error feed back procedure worked well. I could get it from 9 dB peak to peak after an auto run to just under 4 dB peak to peak after just one correction using the spreadsheet calculated data points.

Finally managed to get it under 3, more like 2.5 to 2.6 dB peak to peak in less than an hour using the Editor App and Excel.

 

[Pogre]

I am happy to report that the spreadsheet calculated error feed back procedure worked well. I could get it from 9 dB peak to peak after an auto run to just under 4 dB peak to peak after just one correction using the spreadsheet calculated data points.

Finally managed to get it under 3, more like 2.5 to 2.6 dB peak to peak in less than an hour using the Editor App and Excel.

Dina called me the other day to tell me my new speakers will be shipping next Thursday or Friday! You've got me really looking forward to trying this out!

Man, maybe tomorrow I'll do a couple of practice runs instead of trial and error-blundering my way through it with my new stuff. It'd be nice to get them all set up and dialed in in an hour. I'm ready to roll my sleeves up and dig into this. Peng, would you consider sharing your contact details of some type (text, chat, messenger) with me? It was really helpful when Grant (@ATLAudio) was showing me the ropes with the Mini.

I've used excel a little bit when I was running the meat shop, but have never started one from scratch or anything. Will I need a regular pc or will a tablet work? Gotta get ratbuddyssey yet too...

 

[PENG]

Dina called me the other day to tell me my new speakers will be shipping next Thursday or Friday! You've got me really looking forward to trying this out!

Man, maybe tomorrow I'll do a couple of practice runs instead of trial and error-blundering my way through it with my new stuff. It'd be nice to get them all set up and dialed in in an hour. I'm ready to roll my sleeves up and dig into this. Peng, would you consider sharing your contact details of some type (text, chat, messenger) with me? It was really helpful when Grant (@ATLAudio) was showing me the ropes with the Mini.

I've used excel a little bit when I was running the meat shop, but have never started one from scratch or anything. Will I need a regular pc or will a tablet work? Gotta get ratbuddyssey yet too...

I can send you my Excel sheet and a simple procedure.

There is a limit though, it look just one single trial to get it within about 3 dB peak-peak from 20 to 120 Hz but to get to within 2.6 dB peak-peak, that is +/- 1.3 dB, it took several more trials.

You already got yours to +/- 1.5 dB but I can have the same now without hooking up the mini. Also, we both know it's just an exercise, and have fun doing it. As for sound quality, I don't think anyone can tell the difference between +/- 1.3 dB and +/- 3.5 dB 20 to 150 Hz.

 

[shkumar4963]

I know you did explain it in the video, but I don't totally understand why moving the mic around in my room would not reveal any such "messed up" effects either, I thought it would, though the results wouldn't be the same as the way you did it, and I don't feel so bad now, knowing that even Gene had questions about it. I will do more reading and think harder of what you are trying to say/explain, and hopefully understand the point eventually. For now, I am seeing that no matter where I placed my omni-directional mic with a left, right, above and below by10 to 12 inches between each positions, that is, up to 24 inches between the left most/right, just as an example, the Reference FR always look much flatter than the "Off" FR curves, so I find it hard to say Audyssey would create something bad or erratic outside of the MMP.



When I did my 9 position listening bubble, or window (I know you are telling me that the way I did it really didn't represent a "listening window"). I limited the range to 200 Hz as you can see the ref vs off FR curves in post#48. So now that you mentioned below 100 Hz, are you saying that at least for below 100 Hz, the way I did it would still be fine?



I understood your (and of course Floyd's) point, but as Jon AA sort of alluded to, I also believe Floyd's (and may be yours too) point might have been occasionally and slightly taken out of context, in the sense that Floyd, you guys may not really be saying that it EQ above the transition frequency would not work, but that it could not be made to work, because the input data collected by the mic in room wouldn't be reliable at all above the transition frequency. I know this sounds confusing, but I tried..

Now, back to what I said jokingly about "feed back loop", it was actually meant to be a bit like what Jon said, but he expressed more or less the same in a clearer way. That is, if we "feed" the actual measured results back to Audyssey, by way of manually entering additional data entry points based on the actual measured in room response to alter the target curve, it would be like feeding back the error signal to the input of the control process (i.e. Audyssey auto room EQ). And I was only referring to doing it up to 200 Hz. I know Jon AA is working on making it work for the higher frequency range, I would be interested in trying to do that too eventually, but for now I just want to do the best I can, for the 20-200 hz range only.

May be I can express myself better with a numerical example. Let's say Audyssey targeted a flat response of 70 dB at 80 Hz, and the measured response was 3 dB above (i.e. 73 dB at 80 Hz as measured) the targeted level, then the error would be +3 dB, so if I enter a cut of 3 dB at 80 Hz, then I am telling, or fooling Audyssey that I want to customize my curve to have a 3 dB dip. If Audyssey is not perfect but is reasonably good, it would try to apply a cut of approximately 3 dB, and then I would get my "flat" response at 80 Hz, though Audyssey would think that I wanted a 3 dB dip instead of "flat". Now if Audyssey cannot achieve close enough to the 70 dB target, but the response did drop to 71.5 dB, then I would make a second attempt by entering a cut of another 1.5 dB; and by repeating this process many times, it would be like adding a "feedback loop" to the Audyssey EQ process, though as you said, and I knew that too, that it is not what Audyssey was designed to work and it would be in their wildest dream that some crazy user like me would so such a thing.



Just want to be clear, are you saying all three failed in the below 200, or even 100 Hz, or just the higher frequencies?

That sounds good to interactively fool Audyssey to get to the ideal curve.

When I try to do that, I find that I can only change the desired Audyssey curve using MultEQ to only about 12 dB per octave slope or something close to it. Do you guys also find this restriction when you try to edit the desired curve?

Sent from my SAMSUNG-SM-G930A using Tapatalk

 

[PENG]

That sounds good to interactively fool Audyssey to get to the ideal curve.

When I try to do that, I find that I can only change the desired Audyssey curve using MultEQ to only about 12 dB per octave slope or something close to it. Do you guys also find this restriction when you try to edit the desired curve?

Sent from my SAMSUNG-SM-G930A using Tapatalk

I assume you meant using MultEQ Editor App, but not sure what you meant by "about 12 dB per octave slope...", can you explain a little more please?

 

[PENG]

I think I am going to stop meddling with the 10 to 100 Hz range for now as I am sure I have reached the point of diminishing return and most likely is now bottle necked by the tolerance/accuracy of the Audyssey mic and the Umik-1 mic such that it would be impossible to get consistent errors less than +/- 1 dB. And, that's assuming Audyssey's FIR, yet minimum phase filters are 100% stable!!

At this point, like Jon AA, I am very surprised Audyssey's filter creation is so incredibly good. Based on the excellent results I was able to achieve, I would say the main reasons why people failed to run Audyssey to really flatten their in room response to within say, 5 to 6 dB peak to peak in the 20-120 Hz range are:

1. Existence of room modes that are difficult or even impossible for such software based systems to solve.
2. User errors, not getting the room quiet enough, not following procedure etc.
3. Really bad speakers and subwoofers.

Finally, I think for those with reasonably good or not too bad rooms, speakers and placement options, the key may be the following:

That is, Audyssey has improved its resolution greatly since the XT32 version, but that may have only helped the most in terms of creating the necessary filters. In terms of collecting accurate data, it may just be not good enough for crazy people like me (Pogre, Jon AA? not saying they are crazy, so don't take me the wrong way..) who want to see better than 3 dB, or even 4 dB peak to peak with 1/12 smoothing, or 6-9 dB without smoothing.

Based on my findings, for those who have rooms that are not too good or bad, like mine that has room modes at 30, 50 (the worst one) and 70 Hz, you will likely be able to use the App to fool Audyssey into creating a non flat target curve to achieve a near flat measured response curve.

Below are my concept, and the cheat, figures used are real numbers but for my room only:

- Audyssey detected the subwoofer's in room response and applies a "flat" target curve accordingly.
- Audyssey collected the data from the 8 mic positions, and created the necessary filters.
- Audyssey's filters "corrected" the response between 20 to 120 Hz from a measured response curve of 14.3 dB peak to peak to 9.3 dB with no smoothing.

So Audyssey was able to improved the response by 5 dB peak to peak in the 20-100 Hz range, without any editing but I would need an additional 5 dB improvement to get to my goal of < 3 to 4.0 dB peak to peak, that is, +/- 2 dB consistently. That seems like a reasonable goal as we know between the two mics (Audyssey and Umik-1), we probably shouldn't expect a combined tolerance/accuracy of better than +/- 0.5 dB anyway.

Still, I want to get 5 dB (peak to peak) flatter response. Since we already know a "Flat" target curve would only get me a non flat (9.3 peak to peak) measured FR curve, that means I have to set a "Non Flat" target curve for Audyssey to aim, in order to have any chance to get me a flatter measured FR curve.

Cheat steps:

1. Use the REW FR graph using the "Reference" target curve, and set up an Excel (or any) spread sheet.
2. Enter the point by point SPL for the 20 to 120 Hz range, ymmv based on your curve, mine is 13-103 Hz.
3. Set a target SPL, based on my visual inspection, I selected 68 dB, so as to end up mostly cuts are needed.
4. Enter the formula to calculate the adjustment for each point on the ref curve to hit the targeted 68 dB.
5. Use Ratbuddyssey to enter the manual adjustment required.

In my first attempt, I made 20 adjustments, that actually got me on target of resulting in 4 dB peak to peak with no smoothing, and 3.15 dB with 1/12 smoothing.

In my second attempt, I ended up with 42 adjustments, and the final curve has the unexpectedly good 3.91 dB peak to peak and peak and 2.2 dB with 1/12 smoothing.

So to save time and not worry about using too many filters that are free anyway, I would say if you want to save time, use at last 40 points on the curve between 15 and 120 Hz (or as needed), again, yours may vary, depending on your in room response curve.

Now I am very happy with +/- 1.1 dB 1/12 smoothing, and still within +/- 1.5 dB if extended to 18 Hz. Next I will do the same to the 100 to 200 Hz range but I am in no hurry for that. Time to start working on my F5 class A amp kit.

 

[PENG]

Final graph for the +/- 1.1 dB, 20-100 Hz: