This is a guide to taking simple quasi-anechoic measurements for the least hardware expense possible, using Room EQ Wizard (
donate!) and VituixCAD (
donate!). This way, I can also stop repeating myself every time someone asks me how I do it!
Warning: this guide will be wordy. My hope is that even a beginner can learn to do quasi-anechoic speaker measurements this way, so I apologize if I repeat myself or state some obvious things. I started measuring speakers with absolutely no engineering background and barely any knowledge of acoustics, and I wished there'd been a wordy guide for me starting out.
I'd like to start by acknowledging the late Jeff Bagby, whose
white paper on quasi-anechoic measurements is basically how I got started; much of what's in this guide is essentially just a 'translation' for REW. Of course, Dr Toole's
book was invaluable for the initial inspiration and teaching me how to interpret that data. To Amir for providing a platform to emphasize speaker measurements, as well as Stereophile, Soundstage Network, Erin/hardisj, and others who provide valuable sources of speaker measurements that I've often used to compare my data with.
This guide will be divided into 6 parts. How many you read depends on how thorough you want to be with your measurements:
- Introduction to quasi-anechoic measurements
- Setup and gear
- On-axis measurement (sans low bass)
- Nearfield bass measurements
- Off-axis measurements
- Create a full shebang spinorama
Each part will build on the previous ones, so you don't have to the full guide. If you're already familiar with the ideas behind quasi-anechoic measurements and just want to know how to do them in REW, you can skip the 'On-axis measurement' section.
Please give this a try! Even if you just perform a single on-axis measurement without bass, that's already a lot more useful than most of the speaker information available on the web.
Finally, please keep in mind this is just one way of doing things, lots of which I've learned through trial and error. I am open to feedback =]
Update 4/12/21: Fixed some typos, reworded some bits for clarity.
Update 5/11/21: Fixed some more typos, reworded more bits for clarity.
Update 5/20/21: Added reminder to make sure sample rates for input and output device match (should be 48 kHz with Umik-1)
1) Intro
The best thing you can do is send your speaker to Amir or Erin for testing with the Klippel NFS. But if you can't do that for whatever reason, creating quasi-anechoic measurements can help contribute to the pool of valuable speaker data. Even a single on-axis measurement can be extremely useful.
Quasi-anechoic measurements are basically a way to take a speaker measurement indoors or outdoors and ignore the influence of walls (including the ceiling and floor). The gist of it is to make a sine sweep in REW and truncate the impulse response such that REW only includes the data from right before the first major reflection 'hits' the microphone (it's much easier to do than it sounds!).
View attachment 120852
This is called 'gating' or 'time-windowing' the impulse response. In doing so, you lose some resolution, which is most apparent at low frequencies, and the data becomes completely invalid for the bass (usually below 100-200 Hz). The wider the gate, the higher the resolution of your data. A 5ms gate, such as is used by Stereophile, will give you a resolution and lowest valid frequency of 200Hz. My measurements are typically done at 6.5ms, which gives me a resolution of 154Hz. The resolution calculation is 1/[window in seconds], so 1/0.0065, though REW will let you know too.
To make up for the lack of resolution at lower frequencies, we can take super-nearfield measurements of the speaker's bass components (woofers, ports, and passive radiators), and simulate the far-field bass response from it. (Another common, even more reliable method for bass measurements is the
ground-plane method, but that requires an ample amount of space, so I've never really used it).
With a bit of care and trial-and-error, you can get results that greatly approximate those made in an anechoic chamber or with the Klippel NFS. For some validation of the method, and an idea of what you can expect, here are some examples of my own measurements compared to anechoic sources.
JBL HDI-1600 (vs Amir's NFS):
View attachment 120854
View attachment 120855
D&D 8C (vs Erin/hardisj's NFS):
View attachment 120856
Focal Chora 806 vs Soundstage Network's at the NRC anechoic chamber:
View attachment 120857
The Spinorama/
CTA-2034A standard says that a ±1.5dB measurement agreement for the same speaker is considered 'good'. You can see the above measurements are very close to that, despite measuring different test units.
Note that this does not mean the quasi-anechoic method is as
accurate as an anechoic chamber or Klippel NFS. In particular, resolution in the low mids pales in comparison, which means narrow resonances may be obscured partially or entirely. But the data is still very useful for determining trends in tonality and can become effectively equivalent to anechoic ones by the upper mids.
2) Setup and gear
Here's what you'll need:
- Room EQ Wizard. This guide was written with beta Version 5.20 RC6. As of writing this guide there are a several important features in the betas not available in the 'stable' release that is currently available on the REW website (V5.19). In my experience, the betas are extremely stable for the type of work we're doing
- (If splicing nearfield bass) The Jeff Bagby Diffraction and Boundary Simulator, for adjusting nearfield bass measurements to match farfield results. This requires Excel; I've not tried it on Google Sheets or other spreadsheet software.
- (If doing full spinoramas) VituixCAD (version 2.0.65.0 was used for this guide). It will automatically create a spinorama once provided with enough horizontal and vertical off axis measurements. It can also adjust nearfield bass measurements, but I prefer the simplicity of the Bagby spreadsheet.
- A MiniDSP Umik-1 or other flat measurement microphone. If you don't already have one, I'd highly recommend getting a calibrated Umik-1 from Cross Spectrum Labs for extra accuracy. It only costs a few bucks more than ordering one directly from MiniDSP ($110+ shipping). It's not necessary, but increases accuracy in the upper treble and lower bass and adds peace of mind.
- A microphone stand. It just needs to be thin so as to be minimally reflective. I use something like this, about 20 bucks.
- A sturdy way to elevate speakers far off the ground, preferably 5+ feet, but as far from surfaces as you can manage. I've typically simply placed my speaker stand on top of a table. The sturdiness of the speaker stand is particularly important if you want to do vertical measurements. I'm currently using this.
- (If doing off-axis measurements) You'll need some kind of turntable to place your stand on. I use this and label it with angles in 5 degree increments. For added security, especially for vertical measurements, I highly recommend getting a rachet or cam strap to secure the speaker while it's off-balance. I use one or two of these.
- Open space. If measuring indoors at 1m — sufficient distance for most bookshelf speakers, in my experience — you'll want the closest wall (including the floor and ceiling) to be about 1.5+ m (5+ feet) to match the time window and resolution I've used in most of my measurements (6.5ms). You'll also want to move all furniture out of that 1.5 foot radius — or as far as possible — but small objects shouldn't cause much of a problem. If you have low ceilings and can't measure outdoors, you might have to settle for a smaller gate or measuring at less than 1m.
The open space is key. Again, the greater the time difference between when direct sound hits the microphone and the first big reflection hits the microphone, the more resolution your data will have, and the lower the frequency your measurement will be valid to.
When setting up your speaker on the stand, it should look something like this (taken from the
CTA-2034A standard):
View attachment 120858
It is important to make the edge of the speaker stand as flush as possible with the speaker's baffle, as otherwise the setup can introduce minor reflections that might look like resonances. And again, I'm using a 1-meter distance, rather than the 2m the spinorama standard technically asks for, in order to increase the available time window.
Neither 1m nor 2m are magic numbers, by the way. For horizontal measurements of small speakers, simply being 2-3x the baffle width is usually enough. For a single on-axis measurement of a small speaker, you might get away with less than 2 feet. Experiment and see how the response changes at different distances, and find the best compromise for your space. Vertical polar measurements will be the most affected by short distances, so I would try to keep at least 1 meter for those for most speakers.
Don't sweat your setup too much. It doesn't need to be too fancy. This is what I used for the JBL HDI-1600 measurements above (set up for vertical measurements):
View attachment 120861
The important thing is to simply minimize reflections enough to keep your data sufficiently clean to be useful, which you can readily assess from the resulting frequency and impulse response. If the impulse response looks messy or the frequency response looks unexpectedly 'squiggly', try to move stuff around to make it as clean as possible, then remeasure. It'll take some trial and error, but again, don't sweat it too much. Perfect is the enemy of good.
One more note: make sure that the sample rate for your input and output devices are the same (Thanks for the reminder
@sweetchaos). The Umik-1 can only operate at 48kHz for example, so you'll want your audio output to be at 48kHz as well. Many devices will default to 44.1 kHz and using a different sample rate can have a slight effect on the highest frequencies in my experience. Using a higher sample rate won't improve accuracy, per REW documentation.
On Windows 10, you can do this by going to Sound Settings> Sound Control Panel, tapping on your playback device's properties, and then changing the sample rate in the 'advanced' tab.
View attachment 130931
You should also make sure any spatial audio effects and the like are turned off.
3) The On-Axis measurement (sans bass)
The most basic quasi-anechoic measurement you can do is a simple on-axis sweep.
It's way easier and faster to perform, say, a single on-axis quasi-anechoic measurement (or even a few horizontal off-axis angles), than to do a full vertical and horizontal spinorama with nearfield bass spliced in. In fact, if you can position the speaker fast enough, it only takes a few minutes to do.
As noted earlier, creating open space around the speaker is key and your setup will likely take the most time in this whole process. Before even making a quasi-anechoic measurement, simply moving your speaker away from walls and measuring from closer — thereby minimizing the 'loudness' of reflections — cleans up the data a lot.
To illustrate this effect, here is an old measurement of the Buchardt A500. This is an on-axis measurement taken as a single sweep from my listening position 3m (~10ft) away:
View attachment 120862
This doesn't tell us much about the speaker's direct sound.
Now here is another measurement taken from just 1 meter, after repositioning the speaker such that it is 5+ feet from every wall, including the floor:
View attachment 120863
The highs are much cleaner now, and we have a better idea of the speaker's sound, but this is still not terribly useful. Next, I'll show the
exact same measurement file you see above, except with a gate or time window applied. Note that this was not a separate sweep, I am simply modifying how REW interprets the same file:
View attachment 120866
That's more like it! Although we lost the bass response, we have now removed the 'noise' of the room and have something that tells us something much more useful about the "true" direct sound of the speaker.
Here's how you do it.
Again, position your speaker as far away from walls as possible. Make the speaker's baffle flush with the edge of its stand. Aim your microphone at the speaker's reference axis; check the manual, but if not stated, it's usually the tweeter or midway between the tweeter and woofer. If you're using a boom microphone, try to keep the arm extended such that the microphone is far from the 'stem' to minimize reflections near the microphone.
View attachment 120868
(Ideally, the boom would be in line with the microphone, but I couldn't get it high enough in this case).
Then just take a regular sweep measurement in REW. I assume most of you know how to do this, but it can be done from the 'Measure' button on the upper left (shortcut: Ctrl+M). These are my usual settings:
View attachment 120869
(Ignore the output and input settings, as the microphone wasn't connected when I took this screenshot).
There is one important setting in the 'measure' window that you should keep in mind for doing off-axis measurements later. By default, REW sets t=0 at the IR peak, but this causes problems once you go more than 90 degrees off-axis (basically, the reflection off a wall might be louder/have a higher IR peak than the direct sound). So it's better to set it to t=0 at IR start. The resulting FR should be the same.
View attachment 120870
Now tap start (or press the spacebar), and once the sweep is complete you have all the data you need!
From here, we just need to change the way REW interprets the data to get our quasi-anechoic measurement.
Head over to 'Impulse' tab, and make sure you're in the percentage view on the upper left. The impulse response shows us the same FR data we just captured from a time perspective.
View attachment 120871
You should now see something like this:
View attachment 120872
See that blip right before 7ms, and how the data is all messy after that? That is where the first reflection hits the mic (each subsequent 'blip is another reflection). Were going to remove those blips from our data. You may need to finagle a bit with the zoom controls on the upper left and bottom right corners to get a good view of the blips:
View attachment 120873
If you're measuring outdoors, you might not see such pronounced reflections. That's fine; don't worry about it too much, as we can always adjust the gate later.
Now tap on 'IR Windows' at the top of the REW window. Set the 'right window' to a time just before your first reflection. In my case, I set it to 6.5ms. The left window is usually not very important, but if measuring outdoors, it may help to shorten it to about 2-5ms to prevent loud sounds from contaminating the data.
View attachment 120874
If you're doing off-axis measurements, it's good to leave yourself a little 'slack' between your window and the first reflection, as sometimes distances change a little bit as you're rotating the speaker. Hence me using 6.5ms even though I could stretch the window a little higher.
As noted earlier, the longer you have before the first reflection hits the microphone, the more resolution you have in your data, and the lower the frequency your data is accurate to.
REW will tell you what the frequency resolution of your measurement is, which will also be the lowest frequency the data is useful to. As shown above, 6.5 ms gives a resolution of 154 Hz. If possible, try to get at least a 5ms gate (this is what Stereophile uses, for reference, although they measure from a further distance), which has a resolution of 200Hz. Still, even a smaller gate can be useful, just know you'll have lower resolution.
And that's basically it. Once you tap on 'Apply Windows' you should now see a cleaner frequency response. You can also get a live view of the changes caused by changing the time window by dragging the green 'R' marker at the top of the Impulse tab.
Then return to the 'All SPL' tab and you should see your new gated measurement.
Some miscellaneous notes:
- Make sure REW is set to a reasonable scaling to get a useful view of data. It's easy to obscure flaws in the frequency response with very tall scaling. If you tap on 'Limits' in the All SPL window, the SPL Top and Bottom should be a 50dB difference in most cases.
- For a better way to ensure consistent scaling when sharing your frequency response, I recommend using REW's built-in 'Capture' button on the upper left. Under 'graph aspect ratio,' select 25 dB/decade. This is technically the aspect ratio defined by the spinorama/CTA-2034A standard too, although not even Harman uses it most of the time. The good thing about using this method is that even if you set different vertical limits than the usual 50dB, your frequency response will export at the same scaling.
- 1/24 is my preferred smoothing.
- You can make your frequency response dashed or dotted by tapping 'Controls' and then 'Trace Options.'
- As we're not measuring sensitivity for this guide, SPL choice isn't terribly important. 85dB @1m is a reasonable SPL, but I used 75dB for a long time to not annoy neighbors. It matters most when using DSP speakers whose frequency response might change (compress) significantly with SPL level.
- Sometimes there are objects that cause unexpected reflections. Others matter a lot less than you'd think (like your own body, sometimes!). You can usually tell if something is amiss by how 'messy' the impulse response looks, or if the frequency response looks unexpectedly wiggly. Again, trial and error. Mess around with positioning and settings until you find something that works consistently.
Phew. I know that was a lot, but you should see it's really not all that difficult. Hopefully, this will get you started!