A Newly Installed Loudspeaker System
As stated in Part I of this blog, our friendly SI had already done some proof of concept work in demonstrating the virtues of a loudspeaker that the owner agreed was an improvement over the existing system. The value proposition was made and the owner signed on the dotted line. A new steerable loudspeaker system was installed.
In my world that simply means more data that can be measured to compare to our room simulation (model). Based on the results of the models room decay data vs. the measured data, I also decided to get more "physical data" on the room. More on that later in a discussion of energy scattering. In acquiring Room Impulse Response (RIR) data for the newly installed system, I had a chance to audition the new loudspeaker in the space, and then take it back home by way of RIR, for the purpose of convolution. Like the audio files in Part I, I've convolved RIR's with dry speech to compare the existing or now "old system" with the "new system." For reference, I also include the column loudspeaker from the initial test
For reference, below is the model showing the three receive locations (R01, R02, and R03), with R01 being farthest from the loudspeakers. You might also notice the addition of the digital steerable (newly installed) loudspeaker above the other shown (test) loudspeakers. The SI steered and tuned the loudspeaker. Below is a comparison of the new loudspeaker ("Steerable Column"), the "Existing" or "Old" speakers, and the test "Column" speaker. I've convolved the RIR's with dry speech. The files are for R01, which is in the farthest position from the loudspeakers.
What did you hear from auditioning the above speakers "in the room?" I think we can all agree that the "old system" is unintelligible and the sound quality is poor. The column speaker used in my initial measurements has what would "score" as good speech intelligibility. Remember, this is a gymnasium withe a 3.5s room decay time. The "steerable column" that was recently installed also has good intelligibility, considering the space, and some might say it has possibly better speech "quality", in that it has more upper octave energy. So, does it have better speech quality or speech intelligibility? Let's look at the STI (IEC 60668-16).
First lets look at the existing "Old" speaker system, then the Column speaker, and finally the Steerable Column speaker, all at position R01.
Now go back and listen to the column and steerable convolved speech files again. In terms of being able to understand the words spoken, they are really equals, and to me ears the column does sound a bit more intelligible. One thing to consider here is that the column was unequalized, while the steerable column was equalized. The added high frequency content of the steerable might be perceived as improving the quality, but it doesn't improve the intelligibility. What were talking about here is whether or not you can understand what is said, and to what degree.
Adjusting The Model
We started with one model, and from that we acquired a statistical RT for the room, and we compared that to the measured RT or room decay. Given that the room is an almost ideal room for trusting the statistical RT (Eyring Equation), given all the surfaces have an absorption coefficient of well below 0.3, we would expect the modeled v. measured to be very similar. For the most part that proved true (see Part I of this blog), but it wasn't very accurate for all octave bands of energy, specifically the lowest and highest octaves of energy. This is why we measure whenever possible! We want greater accuracy! So, what to do now? Adjust the model!
We might consider this as much art as it is science. Maybe a better explanation is that we can use empirical data and common sense to make our changes to the model. The first thing you have to consider is what acoustical properties did you assign the surfaces in your model. I used what is labeled as hardwood court, 2x5/8 gypboard, and painted masonry. Truth is that I don't really know what is behind the gypboard walls in terms of how it is constructed, and it would likely vary for exterior walls verses interior walls, which this room has both. The painted masonry, isn't painted brick, it was the closest thing our SI could find to put in the model. The material was actually concrete masonry unit (CMU). Well maybe those are one in the same? The bottom line is the room decay is much longer at the lower octaves than what the model shows! Looking at the data for 2x5/8 gypboard, we find it shows absorption (transmission) by diaphragmatic action in the lower octaves, so that seems like a logical place to make an adjustment.
Making minor adjustments to the surfaces provide us with a "better model." Directly below is the room decay for the adjusted model, and below that, is the room decay for the measured room.
At this point we could stop refining our model and start inputting acoustical treatment into the model. The treatment will help improve intelligibility for both reinforced and non-reinforced speech, and help control the ambient room noise levels that will interfere with speech intelligibility. My take is that is stopping short of what needs to be done! Why? I would agree that you could at this point take the above data, use the Eyring and even the Sabine equations to determine the sabins needed to attain a goal room decay, but that wouldn't be the best solution when considering the reinforced sound. The reason it isn't the best solution for the reinforced sound is because the SI has installed a highly directional loudspeaker, and a cursory look at the room decal in detail (ETC or Reflectogram) shows us that there are some high level reflections that arrive back at the listener area at deltas in time and level that are cause for lowering speech intelligibility and speech quality. In other words, high level reflections are cause for low intelligibility scores.