How to measure your room with REW

Two things, and neither substitutes for the other: a measurement microphone with a known calibration, and REW itself. The software is free. The microphone is the only thing you have to buy.

• A measurement microphone. The miniDSP UMIK-1 is the default choice because it plugs in over USB and needs no audio interface and no preamp, and because miniDSP publishes a calibration file for your individual microphone, matched to the serial number printed on it. XLR capsules cost less but need an interface with phantom power, and usually ship with a generic curve for the model rather than a file for the one you own.
• REW, from roomeqwizard.com. Free, and there are builds for Windows, macOS and Linux. It is a Java application, so it may ask you to install a runtime the first time.
• A tripod or a mic stand. Not a hand, and not the arm of the sofa.

Five minutes here decides whether the measurement means anything. Four of the six mistakes at the end of this guide are made in this step.

1. 1.Choose the devices. In the soundcard preferences, set the input to the measurement microphone and the output to whatever feeds your amplifier. Play a sweep at a low level once and confirm the sound comes out of the speakers you intend to measure.
2. 2.Load the calibration file. Download it from miniDSP using the serial number on the microphone body, then load it in the mic and meter preferences. A UMIK-1 without its file reads several decibels wrong at both ends of the range.
3. 3.Load the right angle. The UMIK-1 ships with two files, one for 0 degrees and one for 90 degrees. Pointing the microphone at the ceiling means the 90 degree file. Mixing the two up puts a broad error across the top two octaves, and it looks exactly like a treble problem.
4. 4.Set the level. Run the level check and aim for a sweep around -12 dBFS. Too quiet buries the low end in noise, too loud clips and produces harmonic distortion that shows up as detail that is not there.
5. 5.Calibrate the SPL reading if you have a meter. Optional for this site, because the analysis aligns your curve to a target before comparing anything, but it tells you whether you measured at a sensible listening level.

Microphone position matters more than any setting in the software. The measurement describes the point in space where the capsule was, so put it where your head is.

• Listening position, ear height, with you out of the chair. Not the top of the sofa back, not the coffee table.
• On a stand, pointing at the ceiling if you loaded the 90 degree calibration file. Held in a hand, your arm and body become a reflecting surface 20 cm from the capsule.
• Clear of cushions and headrests. A soft surface right behind the microphone absorbs the top end and you will read it as a dull room.
• Same spot for the left and the right sweep. If the microphone moves between them, the channel difference you measure is partly the move.

Measure the left speaker on its own, then the right speaker on its own, and name them so you can tell them apart later. Both channels playing together is a different and less useful measurement: the two arrivals interfere, and a level or placement difference between your speakers hides inside the sum instead of showing up as the difference it is.

• Sweep length: 256k or 512k. A longer sweep buys signal to noise, which is what the bass and the RT60 numbers depend on. It costs you a few more seconds per measurement.
• Range: the full 20 Hz to 20 kHz. Exporting a narrow slice leaves the analysis with too little to work with, and this site rejects a file spanning less than a couple of octaves.
• Repeat each channel once and compare. Two sweeps of the same speaker from the same spot should lie on top of each other. If they do not, something moved or something was making noise.

Two exports, and the second one is optional but changes what the report can tell you.

1. 1.The frequency response. With the measurement selected and smoothing set to None: File, then Export, then Export measurement as text. Do it once for the left measurement and once for the right. Each file holds a frequency, a level and a phase angle per line.
2. 2.The decay. Open the RT60 tab, then Export, then Export RT60 data as text. This one holds a handful of lines, one per third-octave band, with the decay times side by side.

Both are plain text you can open in any editor and read. The frequency response is what the analysis here needs; the RT60 export is what lets it size an absorption package instead of only telling you where panels should go.

• The calibration file was never loaded. Reads several decibels wrong at both ends and invents a problem at each extreme.
• The 0 degree file was loaded with the microphone pointing at the ceiling. Broad error across the top two octaves that looks like a dull room.
• Smoothing was left on. The narrow peaks you are hunting are averaged away before you ever see them.
• Both speakers were measured together. Left and right cancel each other in places, and the one problem that is genuinely easy to fix becomes invisible.
• The microphone sat on the sofa back rather than at ear height. Reads the response half a metre from where you listen, which above a few hundred hertz is a different room.
• Something in the room was making noise. Raises the noise floor, and the RT60 bands go empty first.

None of these are subtle once you know them, and each leaves a specific shape on the graph. The trouble is that three of them leave a shape that looks like a room problem, and people treat them by buying panels.