The Worst Way: In-Room, Unaveraged, Unsmoothed

Pros: Easy, cheap
Cons: Marginally useful

The easiest possible way to measure a speaker-or really, approximate measuring a speaker-is to do as I described in the introduction: put a microphone in front of it, run a frequency-response sweep or pink noise through it, and measure the result with a real-time spectrum analyzer, such as TrueRTA or OmniMic. Honestly, this is what I did the first time I tried to measure the response of my audio system more than 20 years ago, using an old-school AudioControl dedicated analyzer with an LED display. In most cases, this measurement is usually done at the maximum resolution of the analyzer, typically 1/24-octave, and it's usually done only on-axis.

The problem with this technique is that it doesn't accomplish the goals of good speaker measurement. It doesn't isolate the speaker's response from the effects of room acoustics, or from reflections of sound off nearby objects. It doesn't let you separately analyze the on- and off-axis response, because even though your measurement is primarily on-axis, you're getting some of the off-axis response, too, after it's bounced around the room a bit. This is true even if you move the speaker and mic out into the middle of the room.

Does this mean that in-room measurements are of no use at all? Absolutely not. Let's consider some ways to improve them.

First, measuring in-room at 1 meter and 1/24-octave resolution doesn't tell you much about how a speaker sounds. As you can see in the graph at right with the purple trace, which is the F208 measured exactly as described above, the unsmoothed in-room measurement gives you a frequency response curve with a lot of hash that obscures what the speaker's actually doing. It's sort of like trying to judge a model's looks solely by viewing her skin through a magnifying glass-you see the details but you miss the big picture.

If you compare a 1/24-octave in-room measurement with an anechoic or quasi-anechoic measurement, you'll see that the F208 isn't producing that hash. So where's it coming from? In this case, it's caused by reflections from objects in the room, which have nothing to do with the speaker's performance. A lesser speaker might produce some hash on its own, but with an in-room, ungated measurement, you have no way of knowing if the hash is coming from the speaker or from room reflections. As I said in the intro, it's like auditioning a singer while a chainsaw's running in the same room.

If you smooth the curve by cutting the resolution down to 1/12-octave (at the very least) or even 1/6-octave (seen in the graph at right with the cyan trace) or 1/3-octave (the graph with the orange trace), you start to get a picture of the speaker's tonal balance. Get rid of all that hash and the major characteristics of the speaker's response and tonal balance-the things you can hear-become apparent. The measurement starts to look more like the one we did in the anechoic chamber.

You still have a couple of problems, though. Worst is that the room modes-i.e., the natural bass resonances of your room, which occur below the Schroeder frequency-are showing up in your measurement. This is also easy to fix.

If you play a bass tone, or even a tune with a consistent bass line, and walk around the room, you'll hear the sound of the bass change quite a bit. That's because the bass waves reinforce each other in certain places and cancel in other places. You can take advantage of this by simply taking measurements in several places in the room, then averaging those measurements. Most of those big peaks and dips you've been measuring in the bass will suddenly disappear.

Averaging measurements from several positions has another benefit: It gives you a good mix of on-axis and off-axis sound, approximating the listening window curve I discussed previously.