OK, I've had a lot of requests for the procedure I'm using to use XT32 as "
Speaker Correction" at high frequencies without all the mistakes "
Room Correction" can cause at high frequencies when based upon steady state in-room measurements at the listening position (which is basically what Audyssey does). I do plan on doing a more formal write-up on the procedure explaining the science behind why one would want to do something like this with references, etc, that would hopefully answer many questions ahead of time, but that may be a while. I want to spend more time experimenting, refining the process and verifying some things. But I know people are anxious to try it so here are the Cliff's Notes of the process to get people started. So this is more of a "leading the horses to water" rather than an all encompassing idiot-proof guide.
1) Make sure you have a good Auydssey calibration measurement-wise (mic placement was good, individual measurements in Ratbuddy look reasonable, speaker distances, etc, are where they should be). You won't want to be starting from scratch because you think one of your measurements in the original Audyssey file may have been messed up.
2) Measure your speakers. If you can drag them and your AVR outside, that's great, middle of a big room OK, but as-placed in the room is probably the most practical for most people. Keep in mind, as boundaries get closer, the less you can trust data at lower frequencies. Even in the middle of a large room, the floor bounce will wreak havoc on your measurements, so don't even think about doing anything in the lower midrange with these measurements unless you can do it outside with the speaker up in the air.
Make sure you're using a calibrated mic with the correct calibration file. For this I use the 0 degree file with the mic pointed at the tweeter. Use a tape measure to accurately place the mic.
For most speakers, I think 1 meter on axis with the tweeter is probably the best place to take measurements for high frequencies. With larger towers, farther measurements may be advisable but this makes it more difficult to eliminate reflections from the measurements. You want to use gated measurements to eliminate reflections as much as possible. You can look at your impulse response to give you an idea of where the large reflections are coming into the signal, but even the smaller reflections will influence the results somewhat. This is an area where experimentation and using some judgement comes into play. Try different measurement techniques--distances and gating lengths--and compare them until you arrive at a technique you're comfortable with (feel you can trust the measurements--they're repeatable!) in the frequency range you want to fix.
I've ended up using 1m distance and 4ms gating as a standard. I haven't used REW for this in a while, but with Omnimic here are the results you get with 4ms gating:
The pure gated measurement (in black) in Omnimic as you can see flat out stops at 600 Hz, and looses resolution above that for a while (which is why outdoor measurements with longer gating times are needed to actually evaluate a speaker full range). The brown line is "blended mode" which blends the gated measurements at high frequencies with steady-state measurements at lower frequencies. That's the one I use as I think the increased resolution in the 800-1200 Hz range is helpfull and they're both basically identical above that.
Once you've figured out a technique you want to use, it's time to take measurements for real. While a single on-axis measurement is OK to get you started quickly if you just want to test the process and will basically give you the same results with a really, really good speaker, in most cases you're going to want to do a listening window measurement. Many speakers with horns and most coaxial speakers will have anomalies exactly on-axis that disappear in the listening window. Naturally, you don't want to correct those.
For various reasons, I think the full CEA 2034 listening window is a bit bigger than necessary for the purpose of calibrating the direct sound, so I typically do a smaller "calibrating window"--basically on-axis, up/down 5 degrees, left/right 10 degrees. But that's entirely up to you to do what you feel is best for a particular speaker.
Note, for all measurements, make sure
DEQ is OFF! DEQ can be great set to preference when listening, but you don't want it on when you're measuring for calibrations.
The first measurement you want to take, is one with Audyssey switched OFF. Do this only after calibrating though, so that it's level is the same as with it on. Then do the same with Audyssey On. I don't think it really matters when doing this method whether you pick Ref Rolloff #1 or #2, I got in the habit long ago when trying to draw new curves of using Rolloff #2 because it's more linear at the top end. But whatever you pick, stick with it.
3) Calibrate! Once you've taken the measurements, you've done the hard part. Now you can finish up at your computer.
First look at the curve with Audyssey off to see what your speaker naturally does and make some decisions on how to fix it.
There's an example (different speaker than the above BTW). The first thing you need to decide is your target. I'd suggest for most speakers simply aiming at a flat response is going to give the results that will be preferred by most people (we'll talk about variations later). If aiming for a flat line, you need to decide where to put it.
For this speaker I decided since it has a nice flat and smooth area from about 1100 to 2000 Hz that averages exactly 80 db, that would be the target.
Unfortunately, with Audyssey at high frequencies, it's all or nothing. If you have a speaker that measures perfectly flat with just one flaw (bump or dip), you can't just fix that. It's all or nothing. So while the curve above with Audyssey off is informative about what to do, you can't fix it. You can only fix the Reference curve. With such a speaker that will likely mean "undoing" what Audyssey did instead of simply fixing the one spot.
So, add the reference curve measurement:
Here, the red line is the Reference curve measurement.
First off, you can see how Audyssey can screw up a pretty good speaker at high frequencies. This is a controlled directivity speaker which doesn't have excessive room reflections, and you can see by aiming for a
flat in-room steady state response for most of the curve, it ends up boosting the upper midrange and most of the treble well above flat in the direct sound that will be headed to your ears. That makes the speaker sound excessively bright. One would be much better off limiting the correction of this speaker to bass frequencies than correcting full range with the standard reference curve.
So for this speaker, most of the deviation from the reference curve you type into Ratbuddy will be fixing what Audyssey screwed up.
To identify the new control points, first zoom in on the upper frequencies such that you make the Reference measurement really ugly. I find it easiest to literally print it out and draw your target line on there for reference. Simply put a dot at each "inflection point" along the curve:
(I ended up deleting the bottom few points from that out of the file.) As you can see that is zoomed in such that each line is only 1 db so it looks really bad, but it allows you to be precise in the adjustments needed.
That may look a bit scary--I plan on doing more experiments with applying a little bit of smoothing and using fewer control points. Most of those small deviations won't be audible so there is probably a point of diminishing returns so maybe you don't need to add so many points. Be careful of smoothing though--it can cause undercorrection of the bumps and dips. But if you want to get a super flat curve, this will do it for you. Frankly, I find it impressive XT32 is able to fix such small deviations, even if they're too small to be audible. The first time or two you try it you may want to only try a few points to make sure you're doing everything correctly.
It is my understanding, that there really isn't a downside to adding more points with Audyssey as there may be with adding a bunch of PEQs with other types of filters and processing. XT32 uses more than 10,000 control points to make up the reference curve. So rearranging 5 of them instead of 50 doesn't mean any reduction of processing of the signal (making it more "pure"). Like I said, with this software it's all or nothing, so you might as well do a complete job if you've made it this far.
For each dot location, simply measure the frequency and how far it needs to go to get to your correction line and enter that as a control point in Ratbuddy.
If you do something like the above, you'll end up with a stack of control points as shown here in Ratbuddy:
Results like this are possible:
This shows the Audyssey off in brown, on with the above corrections in black.
Here's the same graph zoomed in:
As you can see, the corrected response is within about +/- 1/2 db from 1100-15KHz. All those graphs have no smoothing at all. Note where the horn "falls off" at the very top end, no effort was made to boost that all the way flat.
The best thing about doing this at high frequencies, is that EQ actually works! At low frequencies, many deviations are caused by room modes, SBIR effects, etc. Some of those can't be fixed by EQ at all, and for some it'll only partially help which can lead to frustration when things don't respond as you tell them to. But at high frequencies, if you follow the above method, Audyssey pretty much does exactly what you tell it--and will nail it on the first try!
There are exceptions though--as noted before, acoustical interferences caused by a bad horn/waveguide (or even a good one above a certain frequency), coaxial speakers and various design flaws of more conventional speakers, etc, may not react well or at all to EQ. If you find a spot in the response that doesn't correct properly, and you've added the control point and its correction properly, it's probably best to just leave it. Throwing gain at something like that isn't going to fix it.
That brings us to other reasons to deviate from a flat line. If your speaker's natural response has a huge dip or massive rolloff at the top end, you may think twice about trying to flatten those things all the way out. If it's bad enough Audyssey may limit you anyway, but you may not want to push them that far, particularly for a speaker with dome tweeters which may have limited headroom. I don't think a few db of boost is the end of the world (Audyssey will surely do it all by itself) as long as you're mindful you will be eating into the speaker's headroom and distortion may become audible a bit sooner. If you know you often listen to the speakers close to the limits of their capabilities, especially to the point of hearing distortion, you want to be really careful about not boosting much if at all. Keep in mind, the proper way to judge how much you're boosting is not the number you put into Ratbuddy, because that's relative to the Reference Curve. Compare the final correction to the response with Audyssey Off to see how much you're actually "boosting."
Other reasons to deviate from the flat line are as simple as personal preference. You may like more of a rolloff at the top end. If so, that's fine--just draw it in and correct the dots to that line instead. You may want a more "laid back" sound in which case you could put a bit of a declining slope into the curve. You may like the "show winning detail!!!!" of a speaker with a big rise the last couple of octaves. That's fine too, the goal is to get the sound you like. This amounts to the second portion of what Kevin Voecks describes in their "Anechoic EQ" where you're using smoother tone-control type adjustments to taste after fixing the "flaws" of the speaker (smoothing out the resonances) with quasi-anechoic data.
The research does tend to point to most people preferring flat direct sound in this frequency range, so that's where I recommend most people start for most speakers. It's a very sensible target that should sound quite good to everybody.
Keep in mind, this does not mean your steady-state in-room measurements will be flat. They should not. They will have a downward slope to them.
This can be clearly seen by in-room measurements of the above result:
In a larger room (with a longer listening distance) the slope will be more, with a speaker that beams at higher frequencies there will be more of a rolloff, etc, but that's OK.
And if the in-room measurements aren't perfectly linear, and they probably won't be with most speakers, that doesn't mean you did anything wrong. That's due to your speaker's directivity index not being linear. That's OK--it's the best you're going to get out of this speaker because you can't fix the directivity issue with EQ. And now you know, if you flatten out the in-room curve with that speaker, you'll be screwing up the direct sound. The direct sound is generally the most important. The above speaker has exceptionally well controlled constant directivity which is why it remains linear in the in-room measurements.
Something else to check after you're done with in-room (steady state at the MLP) is to make sure the high frequency portion blends smoothly with the midrange. If you choose your "target line" poorly, you may see a sort of "step" where the high frequency EQ kicks in if you corrected to a line that is too high or too low. In this case you can simply add or subtract 1 db or whatever the error looks like from all the high frequency control points. Your in-room curve should smoothly go from the point you left the reference curve to high frequencies.
A couple more notes:
This operation is separate from what you do in the bass region. In the bass region, you want to correct the speakers and the room together with steady state in-room measurements--Audyssey does this pretty well when used as intended. Whether you leave the bass flat, add in a "Harman preferred" rise at the bottom, etc, is up your preference. You can try different things on the bass while leaving your high frequency EQ intact by using different copies of the file you made. You can add control points in Ratbuddy out of order. I've found it's a good idea to save the file, open it and save it again before trying to use it though (sometimes it seems to take an iteration or two of that before puts the points in the correct order).
That should be enough to get you up and running. Man, if this was the "Cliff's Notes" version, the full version explaining all the theory and giving references, etc, is really going to take a while!
