Measuring Room Acoustics
Acoustic Measuring Software of decent quality has become very affordable. This has led to many attempting to use it to evaluate their studio or other listening room.
This can be confusing and disappointing. To get the best from any tool some understanding of the principles and operating skill is needed. Measurement is quite a sophisticated activity and the Software has become more and more complex over time. The aim of this article is to offer a distilled basic understanding and the simplest route to getting up and running.
Sophisticated measuring of time based acoustics became available in the 60’s with the invention of Time Delay Spectrometry. For the first time we could properly see and measure sonic reflections on screen. The hardware was expensive until the home PC was harnessed by Acoustisoft with their very affordable ETF software. Still in the game, acoustisoft.com has great walk-through tutorials. FuzzMeasure Pro is a similar product for the Mac. This one is very easy to use, but also sports sophisticated features such as combining measurements in the Time Domain. Room Eq Wizard is very comprehensive, which results in a learning curve. It costs a nice round figure, 0, and it has a highly educational manual. Unfortunately Apple do not implement Java code entirely correctly so REW doesn’t work with multichannel or Firewire interfaces connected to the Mac. However REW works fine using the onboard Line In/Out, whose sound quality is absolutely good enough for the job at hand. There are other apps. ARTA is highly regarded. Smaart is widely used in Live Sound. Studio Six Digital have iApps. There are others, some wildly expensive for no apparent reason.
There are integrated packages which both measure the room and generate corrective Eq filters. Dirac Live. OmniMic. XTZ. ARC2. Trinnov. These systems have become known as Digital Room Correction, perhaps an overstated term. In domestic or pro-sumer situations, i.e. inadequately treated rooms, DRC can be extremely useful. It can make a bad situation workable, while in a good acoustic it will still enhance. An often overlooked and IMO over-riding benefit of DRC is the ability to chose and manipulate a Target/House Curve. On board speaker Eq tends to be crude or non existent. The ability to finely define a full spectrum House Curve has been a game-changer in my experience.
Let’s start with an idealised sound stimulus, i.e. all audio frequencies starting instantaneously and simultaneously. Balloons and starter pistols have been used. Electronic stimulus is obviously more convenient and controllable. So we play a sine sweep from say 10Hz to 20KHz i.e. all audio frequencies of interest stated serially and slowly. While this plays, we record what is happening in the room.
Play and Record, what could possibly go wrong? The subsequent recording will contain all frequencies of interest, PLUS the combined responses of the room and speaker. Magically our software now SUBTRACTS the sweep, leaving us with only the room + speaker response. We now have a map of what will happen to any sound played on those speakers in that room. An Impulse Response. These IR’s are the same as those used in convolution reverbs. Indeed Arjan (Altiverb) explains the IR really well in the first two minutes of this video. Perhaps a picture says it better. Here we see the idealized initial impulse, a single instantaneous spike of 0dB Full Scale, followed by reflections from nearby surfaces. The level and arrival time of these reflections show if they are welcome or not. In critical listening rooms, it is common to see all reflections within the first 20mS suppressed by at least 20dB. Here we see an IR Graph, but the Envelope Time Curve variation is most commonly used due to some advantages.
The frequency response bumps of any reasonable mic are tiny compared to the 30dB anomalies we find in real rooms. Omni is necessary though. Jump in, get a demo of the software and go for it with whatever mic you have at hand. In time, if you develop a liking for measurement, you might want a more suitable mic or software. Measurement mics are omnidirectional small diaphragm condensers. The best of these are optimised to have a very accurate and stable frequency response at all angles of incidence. This requires small capsules which can lead to poor signal to noise ratio. There are a couple which are flat and genuinely omni, and quiet enough for recording. e.g. DPA and EarthWorks. The HF response of all the commonly used cheap mics is dodgy. They are not flat and pointing them in spurious directions doesn’t make them so. This can be entirely fixed by choosing a mic that come with a Calibration file, e.g. this Dayton EMM-6. REW recognizes the new UMIK-1 and the app. automatically becomes Calibrated for both accurate Frequency Response and actual Sound Pressure Level. All of the common affordable measuring mics are designed and Calibrated for simple ON- AXIS use. (Although Dayton also provide very useful 0, 45, and 90 degrees files). So point the mic at the sound source of interest, typically a single speaker running. Exception-For simple broadband level balancing of say 5.1 rigs it is useful to point the mic at the ceiling or floor, whichever is less reflective. The mic will now ignore the HF from all horizontal sources equally badly. Sound Level Meters are very useful in the studio and elsewhere. It is good to get a sense of what SPL numbers actually mean and sound like, e.g. 60dBA, or 90dBC Slow. However I absolutely do not recommend using the mic in any cheap SLM for measurement. The Frequency Response can be limited, unknown, or HF boosted. The self noise is usually awful, sometimes even including DC.
Where to point it
For repeatability I recommend hanging a small pointed weight from the measuring mic. Adjust the length of the thread to so that the weight just barely touches the ground. Mark this spot on the ground using masking tape and label it with a name or number. Obviously use the same name or number when saving each measurement. Perfect recall of mic location, including height.
It is useful to start with a single measurement exactly equidistant from both speakers. Locate the mic there using tape or laser measure, or using by the central null caused when Pink Noise is played with one speaker phase reversed. We can now check if the speakers are delivering equally, bearing in mind of course that the room surroundings at each speaker may differ. In any case a useful place to start. Many of have our own favorite methods of covering the sweet spot. I like to place the mic at the left ear position when measuring the left speaker , then the right ear and speaker. One central pair at the Mix Engineer’s spot. Then single shots within and at the outline of the sweet spot. One way or the other measurements should be taken throughout the listening area which will be actually used, prioritizing the Mix Engineer’s spot unless of course we are in a Home Theatre or such.
I hope that pretty soon all measuring mics will come with both FR and SPL Cal files. This will eliminate the very very common confusion which arises over Cal. This muddle is very understandable for several reasons. For a start many of the manuals and primers place Cal first, as if it were of primary importance. Then we find practicing acousticians stating that for most of our work and in most cases Cal is not at all necessary. On top of that there are three or even four different types of Cal or Correction. Are we sitting comfortably…..
Level Calibration is similar to proper gain staging in recording. The aim here is to achieve a normally decent recording level while adjusting the app meters to read something close to the real life SPL of the measurement sweep. A real SLM can of course be used to Level Calibrate any mic/preamp/software setup. Place the SLM next to your mic. Play Pink noise and adjust the volume to some round figure on the SLM, say 80dB, C or Z, Slow. Now Calibrate the meter in your app to read the same. If you don’t have an SLM, you can still get into the ballpark. Sing a sustained note about 1m from the mic. Use your mic pre gain to achieve about -14dBFS on your Interface or other ‘hard’ meters. Adjust the meter in your measuring app to read the same -14dBFS or 80dB. Uniquely the absolute sensitivity of each UMIK-1 is reported internally to REW, causing the REW SPL meter to show actual levels.
A Microphone Frequency Response Cal file adds and subtracts a few dB’s here and there to directly counteract the measured variations of the actual mic in use. This renders the response effectively flat, of course only when it is pointed the same as when the Cal file was made.
Loopback Correction. Yes there’s yet another Calibration. A report back loop is created between your Output and Input. This can measure any deviations of Frequency Response in your soundcard and thus generate a SoundCard Cal file which can be used to fully correct them. As SoundCard variations are typically a fraction of a dB this strikes me as a bit of a waste of time. LBC has another very significant benefit though. By measuring the time delay in the software and hardware it will allow the software to correct the graphs to read actual times. It is useful to have the initial IR spike at 0, with subsequent reflections reading their delayed time of arrival directly.
So, if you are fully Calibrated, your mic and soundcard will become ruler flat and your graphs will read actual SPL and delay times. Nice, but as were are only really interested in very large variations, such absolute values are absolutely not necessary.
Take a Shot
Remember we are about to perform a simple everyday studio task quite similar to say recording an electric guitar or bass amp.
Establish Playback. Select any output and connect it to a single speaker. Set the sweep to default or say 10-20,000Hz over at least a 10 second period. Start with the volume/monitor controller very low. Hit measure and listen for the sweep. Try a series of sweeps with increasing levels until it becomes a tad uncomfortably loud. Earplugs or sealed cans are a good idea. Watch for clip indicators on your amp or active monitor, particularly at HF. We want a sweep distinctly louder than the background ambient noise. If you have an SLM, if you have performed SPL Cal, let’s try for around 80dB (C/Z and Slow weightings). Expect peaks of up to say 90/95dB.
Now establish the input. Find an input level meter in the software. Tap and mic and you should see action. This tapping should not be audible on the speaker. Now play some measurement sweeps. Adjust the mic pre gain to get somewhere close to full scale on the input meter just like in normal audio recording level setting. Note the default Full Level on the FM Input Meter is 94dB. If you cannot get input or if the level is way off, something in the computer is blocking or boosting. Look into Sound Preferences or Drivers in your OS. Set any available input level faders to 0dB, nominal, or full on. After each trial sweep let’s view an SPL graph. Expect big big peaks and troughs. When happy with with the average levels appearing on these graphs, name and save your first measure.
Our measure can now be viewed in many different ways on different graphs. Consider them as different lenses. The IR itself is not altered in any way by using these views and filters. Play with the controls, Zooming, Dragging, Smoothing, all of it. I have made my settings visible in all the graphs here…… hint hint. It is good to dip into the manual for the software now and then. The REW manual is an excellent read in any case. There are inevitably various little anomalies in the apps. Generally these matters improve with software updates so hopefully some of the following will be fixed. Waterfall settings may not hold when hopping from measure to measure. Scroll wheels or trackpads can be a bit wild so remember that graph extents or limits can be typed in numerically. For Y axis levels try 95dB or 0dBFS for the top to correspond somewhat to the actual peak SPL of the sweep. Let’s say 40-45dB for the bottom to correspond to actual ambient room/machine noise.Smoothing, Zooming, and Frequency Span are all interactive on the X axis. For modal decay work try viewing 1000ms or even longer for untreated rooms, perhaps 500ms for treated. There are controls for ‘windowing’ but this is under the bonnet stuff, i.e. the automatic figures generally work. A full range Frequency Response graph will look crazy at the top end with no smoothing. Use 1/3 Octave or whatever you fancy to get a sense of perspective. This will remove the wild squiggles and show the overall trend or tonality. Flat is not necessarily best. Many of us favour a House/Target listening curve, sloping downwards towards HF. No smoothing at all obviously reveals the most detail. This is useful when zoomed in to narrow ranges of frequency , e.g. the typical 20-300Hz LF range. A reminder, these viewing controls are like those on a microscope. They do not affect the sample.
Frequency Response graphs hardly need an introduction here. But they do not necessarily show the most audible aspects of the acoustic. The ear is well used to adapting to different responses and tonalities. FR is a static measurement. But some software provides Real Time Spectrum Graphs which can be used in conjunction with Pink Noise. This arrangement will give a FR graph which updates every second or so. This can be extremely useful for saying finding an optimum Listening Position. Simply walk about with the mic while viewing the RTA. Look for the most even LF response.
Music is a train of impulses, each of which stimulates the room’s response. In real rooms, sound decays at different rates at different frequencies. LF decaying slowly can make it very hard to hear the pulse of music. HF lingering on can be harsh. There is a whole family of graphs all showing Room Decay in different ways. This is testimony to the fact that the length and spectrum of the room tone is more destructive of fidelity than the more often considered FR variations.
Envelope Time Curve is a variation on viewing the Impulse Response graph directly. Both are very similar in that they show a pretty literal graphic representation of a hypothetical or real single Impulsive spike followed by a train of reflections from all the boundaries. It is useful to Normalise these ETC graphs to 0dBFS and to shift the initial spike to 0mS. Room Tone Decay Time can be derived from the slope of the curve. The absolute level of reflections and their time of arrival after the direct are visible, an indication of how live the room is.
T30 gives a fair indication of how long sound will carry on after the source is stopped. Even better is the specially created Topt in REW. Avoid EDT, T20 and such unless you have a knowing reason to go there. Reverberation Time, RT60, strictly speaking is not defined in small non diffuse rooms. The BBC required their regional studio Control Rooms to have third octave decays within 10% of each other. IMO a wise and powerful qualifier.
The Waterfall or Cumulative Spectral Decay is fairly self explanatory. It gives an easily understandable picture of the spectrum of the decay over time. But note the actual numbers are indicative only, decay times read directly off these graphs won’t be accurate.
There is also the Spectrogram which many favor when experimenting with LF traps.
These decay revealing graphs vividly show how modal resonance rings on. Such modes cause individual bass notes to pop up loudly at particular places in the room, while other notes can be virtually inaudible. These modal peaks and nulls are solidly located in the room. It is very useful to get to know where they are. A modal map if you will. There is confusingly another form of non resonant null caused by SBIR. A single or short series of destructive reflections can make a bass note inaudible. These nulls vary in frequency with different locations of speaker and listener. They appear as a void in the Frequency Response. Sometimes this will be present only for the first say 50ms, then they get filled in by blooming modes close in frequency. Such nulls viewed with no smoothing can be seen to be as deep as 30dB, entirely removing some bass notes, making LF mix decisions very unreliable.
As we are dealing with Decay, it generally starts at 0dBFS if we have normalized, decaying to say -50dB or so, the ambient noise floor. Graph duration and the bottom level of the graph are of course interactive. Adjust them both to fully reveal those modes tailing off into the noise floor of the room
What to expect
Horrible looking graphs! Even the best Pro Studios don’t show their measurements because they are simply way too ugly. It takes an experienced brain to evaluate how a room sounds or errs. A lot more to figure what to do about it. Let’s not expect a health check and a treatment prescription from Software. We see many graphs posted with the question, ‘how does my room look’. The answer is ‘wrong question’ or ‘squiggly’. These are tools, not room doctors.
Impulse Response based measurement is very sophisticated. Simpler tools are sometimes just as useful. Pink Noise and a Real Time Analyser give instant results. Simply watch the Graph while moving the speaker or mic. Sine Waves and an SLM are ideal for precisely finding modal hot spots suitable for LF treatment. REW has these tools built in. A friend Bogic Petrovic has created a really cool set of tone bursts which are ideal for stimulating rooms into ringing. It can be downloaded here. Impulsive Tones can be generated by mouth. Handclaps, Starting Pistols, Pillows, Stopwatches, Balloons, are all useful in the experienced hand. Analysis of the room, by whatever means, can be extremely useful, but it is not mandatory, nor a rite of passage, nor the magic path to audio bliss. A room treatment can often be prescribed by simply looking at it. Listening is helpful! A collection of Reference Tracks can become audio home to the ear and brain. Knowing the sonic signature of these Refs, the response of the room and speakers is immediately obvious to the trained ear. I refer to this type of activity as Active listening. There is intent, learned sonic signature, comparison. On a similar note, we could passively look at our Graphs. Or we could become active invoking physical changes and viewing the sonic results caused by them. e.g. Move a speaker 50mm, view the difference. The following improvement from Red to Blue was accomplished by moving the speaker and listener to optimal positions.
It shows improvement caused by change immediately simply and visually. It would take a long time listening to be so sure of cause and effect.
We measure by playing one speaker, then the other, then both. L, R, L+R.
All of the usual measurement mics perform at their best when pointed directly at the sound source.
The Mac OS does not implement Java correctly. As a result, REW, a Java app, will experience trouble with Firewire and Multichannel Interfaces. But the Mac’s onboard I/O works fine.
Enjoy, DD July 2013