Based on the system EQ settings (and process) that inspired this blog entry a couple of weeks ago, I took some time to do some system measurements to see what kind of pink-noise response the system gives now it’s been tuned “by ear” in our building. This measurement session also set the groundwork for another blog entry concerning the spillover from stage monitors into the rest of our building.
I freely admit that these measurements were taken from mere curiosity and without any specific question in mind, nor were they taken with any specific point to investigate. For this reason I find the measurements and their interpretation so interesting.
Theory of RTA measurement in sound systems
For true reproduction of a pink noise source, one would expect the 1/3-octave bars in the RTA display to all be at the same level – pink noise has the unique property of having equal power per musical octave. I would suggest that this means there is as much power in the range from 20-40Hz as is found from 2000-4000Hz – hence explaining the “flat” response that should be seen on a 1/3-octave RTA such as my iPhone app used for these measurements. Further reading on the theory of pink noise can be found in this wikipedia article on the subject.
A “perfect” sound system
I think the most commonly accepted definition of a “perfect” sound reinforcement or playback system is that it plays back exactly what is fed into it. This definition seems to hold true in most fields, at both consumer and professional levels. The thinking here is that if you play pink noise into the sound system, you should observe an exact replica of that original signal coming out of the speakers. Any tonal (frequency-based) deviation from that original signal will show up on a 1/3-octave RTA display as a peak or a trough relative to the the neighbouring bars.
This is the basis on which EQ’ing/analysing with a pink-noise source is built – we know the source, and in theory we can therefore shape the frequency response of the system to get the height of the bars as even as possible. Having made suitable adjustments to the sound system EQ and speaker choice/positioning, we ought to be able to draw a horizontal line from the top of the lowest frequency bar to the top of the highest, with no one bar falling above or below that line.
Of course the reality of many sound systems, even the ones we would judge to be “excellent” often fall quite short of this aim, due to a combination of physical system/room parameters and interactions, as well as the intentional “voicing” of a system by a human engineer/operator to flatter its usual source material. Humans often like to interpret things in this quite non-scientific way and when this is done well it is considered an artistic addition to the system. When such deviations are present due to the nature of the room or the system itself, or if they are due to an inappropriate “voicing” by a human engineer, these artefacts tend to be considered bad things.
On with the measurements!
Sound mix position
With all the above in mind, I thought it interesting to measure the output of our system when fed a pink noise source, after the system was EQ’d by ear to a series of test-tones, where I intentionally made a perceptive judgement of the apparent loudness of each tone relative to the others. Since these judgements were taken from the sound mix position, I thought this a good location from which to take my first reading, shown here:
Sound mix position (802/302 only)
Now, this graph is interesting. Well okay, *I* find it interesting, even if nobody else does! Overall, it suggests a reasonably even response from 20Hz up to around 2KHz, before the response of the system falls off significantly at around 4KHz, before climbing steadily up to 16KHz. Above this point, either the sound system or the measurement mic seem to show a reduction in output. I’m not terribly concerned about anything above 16KHz because many people cannot hear much above this anyway, and those who can will tend to not be too bothered by a slight reduction in output here.
But what about that dip centered around 4KHz?
A good question. Human hearing is an odd thing, especially in buildings when listening to abstract tones that bear little relation to real music or human voices. Frequencies have different perceived loudness compared to others, even if a perfect sound pressure measurement system shows them to be played at the same frequency. The human ear tends to be most sensitive to mid-range frequencies, from around 1KHz up to around 5-6KHz. This contains much of the intelligibility components of the human voice, so in some ways it makes sense that our ears are tuned to be most sensitive at such crucial frequencies.
BUT: here’s another thing – our hearing sensitivity changes depending on the volume of the sound we’re hearing and/or responding to. Typical speech levels tend to be in the 50-70dB range in normal conversation in a reasonably quiet room, so I chose 65dB(A) as reference sound pressure level for the test-tones as well as for the pink noise analysis, given that it’s towards the top end of the volume range most people find comfortable when listening to reinforced sound. Any louder and people feel like they’re being shouted at. Any quieter and the effects of tonal inconsistencies tend to be perceived as being less of an issue to the average person.
So what has happened here during my EQ session is that I’ve intentionally pulled a chunk out of the system’s sensitivity to these critical mid-frequencies based on the fact that they seemed so much louder to me than the others, despite their measured sound level being within 1 or 2 dB(A) of those I didn’t find so bothersome. On its own, this is usually a bad move for setting up live sound systems for either speech or music reinforcement, so I checked out the new EQ profile by playing some well-known and much-loved music through the system, and found that the EQ curve still worked. Comparing that to the same music played through a curve without the drop in the 2-8KHz frequency range made the music sound rather harsh and shouty, even brittle somehow. So despite a significant issue showing up on paper, I left the new curve in place on the basis that it sounds pretty reasonable with music and speech playback without any further work being required by the mix engineer to make that material sound good.
Phew – so far so good.
Front row position (seats closest to the stage/chancel)
With my curiosity still ablaze, I took another measurement of the system from the front row of our main downstairs space, using the same pink noise signal at the same input level, and leaving the all other controls alone. This will show up any difference between what is heard at the mix position and what is heard at the front row location.
First off, I note that the measured sound levels at this position are really not appreciably different to those experienced at the mix position – 0.3dB as shown between the 10-second measurement periods at each position. Let’s consider for a moment that most people (even many sound engineers) would have a real struggle to reliably notice a 1dB difference between two signal levels. Essentially we’re saying that someone sat at the front of the venue shouldn’t be hearing a signal level difference compared to the same person sitting at the back of the venue, at the mix position. That’s quite astonishing, and shows something of good sound system design that the coverage is so even.
The second thing I notice is about the content of that sound level. The output at frequencies below 2.5KHz shows to be generally slightly lower than measured at the mix position, but frequencies above 2.5KHz are played slightly higher than at the mix position. Thus, despite the overall levels being so evenly matched, from the measured differences one could expect the system will sound tonally quite different in the front row seats compared to the back.
From experience I can tell you: it really does! I now know that if I’m mixing for “hifi sparkle” (with lots of high-frequency detail) as heard at the mix position, the seats in front of me will receive a mix that is unpleasantly biased towards higher-frequencies, sounding more like the brittle and shouty system that the aforementioned measures-flat-on-a-meter EQ method gave us. Ugh.
Centre aisle, front row level (802/302 only)
The sound system as heard by the most important microphone feeding it
An interesting one this. I’ve often observed that I can hear a whole lot of the main sound system output at the main speaking position we use, and I’ve been telling less experienced readers/service leaders for years not to let this fool them into thinking they can speak more quietly and let the sound system do all the work.
Well now I finally have something approximating proof of what I thought I’ve been hearing for all these years:
Central Aisle, Lectern mic position (802/302 only)
Again, there are two things I would note from this graph. The first is that we’re only approximately 1dB(A) down in overall output volume from the sound system at the microphone position than we would be at the back of the church. It’s almost no better than having the microphone directly in front of one or more of the speakers its output is fed into. Gain-before-feedback is a significant area of struggle in our building and with this information I can see why.
Secondly, let’s think about the content of that sound. Again, the output is quite even up to around 3KHz, with less of a relative dip at 4KHz than in either of the measured positions where sound output from the system is desired. In this position we actually don’t want the sound system to be contributing any significant output – and in our current setup we have the very opposite of that desire! A “flat” system response to pink noise would result in a microphone that is already very sensitive to frequencies between 2KHz and 8KHz receiving a lot of its own sound from the system at those very frequencies – leading to vast quantities of feedback if the person speaking into it is delivering less than around 65dB. Sadly for us, many of our less experienced people using the lectern tend to deliver less than 60dB to the mic, so getting them sounding both loud enough and pleasant enough to listen to is a significant challenge.
I hope that rather than bemoaning an ongoing struggle I’ve actually contributed some useful thought and input to the subject that others can either learn from or correct me on. This has been a wonderful learning exercise for me, and I hope the findings can eventually lead to some significant improvements to our systems, our methods and eventually the sound that is heard in our venue.