Sydney Opera House Concert Hall IR

Making the Sydney Opera House Concert Hall IR's
by David Gauger II

The Sydney Opera House (SOH) is one of the world’s most recognizable buildings. Not only does it contain a hall for opera, but there is also a large concert hall (2679 seats), which is of world-class acoustic quality. Unfortunately, it is tough to get IR’s of a world-class hall like the Sydney Opera House concert hall these days unless we think creatively!

The SOH IR’s on this site are derived from acoustic research measurements taken in the Sydney Opera House Concert Hall that have been published on the web. Different kinds of measurements including HATS sweeps, Kemar head sweeps, and MLS Impulse Responses have been collected all over the hall in a systematic fashion. Unfortunately, the original IR files are not only mono, but also extremely noisy, meaning that there is a tremendous amount of hiss present in them. I have cleaned up and repurposed these measurements to create stereo IR's out of them. Instead of pristine scientific recreations of the Sydney Opera House Concert Hall, the IR's should be considered as based on the acoustics of that hall because the raw IR’s were so noisy I had to use much unusual processing to create them. While purists may be appalled, it is my hope that many of us will find them to be musically useful.

Here is the URL of the source data:

http://people.arch.usyd.edu.au/~basse_j/auditoria.html

My guess is that the SOH IR’s published on that site were never intended to be IR’s used for musical reasons in convolution engines. The original files were MLS measurements that someone converted into mono IR’s in Cool Edit. The journey from the data at the above URL to these stereo IR’s is a long one which some may find interesting.

1. The first step was to remove as much hiss as possible without too much damage to the frequency response of the file. I used the Hiss Reduction plug-in in Adobe Audition.

2. The amplitude of the IR at the beginning of the file is loud enough to mask the high noise level present. The end of the file was a different story. I used an audio envelope to ramp the amplitude down at the end of the file to cut the hiss even further during the quietest portions of the reverb tail. Unfortunately, this also cut the liveliness and spacious quality of the end of the reverb tail not to mention the length of the reverb itself. Not to worry: I fixed this in a later step!

3. Note that all the original IR data was collected in mono. To solve this problem I paired the mono IR’s subjectively by ear (and location in the hall) then combined each pair in Audition to make “stereo” (or at least spacious sounding) reverb IR’s in interleaved .wav files. Volume mismatches and other differences were minimized as much as possible by IR selection and processing. (If you think you can improve on these pairings, be my guest! We'll all be better off for it!) Occasionally I had to amplify one channel or the other to make the stereo balance somewhat more even when centered between the speakers. A few times I had to raise the level of the latter part of the reverb tail a dB or 2 to make the late reverb stay centered in the speakers. Processing was used sparingly in this step – only when absolutely necessary.

By the way, for spacious reverb, engineers often want to make sure the reverb is “decorrelated” in the left and right channels. Few things make the reverb more decorrelated than using a different IR altogether, which is what we were forced to do here! I think this contributes significantly to the great sense of space in this IR set.

4. To compensate for the end of the IR’s sounding “chopped off” at the end, I faked a longer reverb tail by reusing the same IR over and over again at successively lower volumes. I used 6 copies of each stereo IR in the stereo mixer part of Audition staggered by about 3/10 of a second or so in time. Then I manually drew in amplitude envelope curves in each copy of the IR so as to hide the chopped off end of the IR with the next copy of the reverb. This had 2 advantages: I could hide the chopped off end, but also I could increase the signal to noise ratio significantly because each succeeding copy of the IR was used at a lower amplitude. When the IR is softer, so is the noise floor! All processing was done at 32 bits to retain the highest quality.

Getting the envelopes of the individual IR’s to be believable was the most time-intensive part of the process. The hard part was to match the timber and amplitude between each succeeding copy of the IR. Obviously, the first IR was the easiest because of the need to retain the initial impact of the “gunshot” so all I did was to ramp out before noise became a problem. I would then bring in only the next copy of the IR, making sure to omit the initial “gunshot” but retaining the next portion of the reverb tail, thereby lengthening the perceived reverb length. If you listen to an IR as a .wav, you may be able to hear a gentle swishing sound as the high frequencies of each copy of the IR decay more quickly than the low ones. Fortunately, your ear doesn't detect this when using the file as an IR.

Here is a screen shot of Audition while in the process of creating these IR’s. Only the portion of file past the initial gunshot but before the hiss level gets too high is used. The actual level envelopes were tuned by hand.

The other thing it does is keep the overall brightness of the reverb a little higher than it would be naturally since were using the portion of the IR just past the initial gunshot where the signal to noise ratio is best. While from a purist’s point of view this is heresy, from a musical point of view, my ear seems to buy the sound when used as an IR with no problems. It is interesting to note if you look at the frequency response of an IR from many Lexicon reverb programs on the web, there is very little decrease in frequency response over the course of the decay. All frequencies seem to decay equally over time. Maybe our ears like this type of sound since Lexicon and others have quite a bit of money creating boxes to do it for us!

So this set of IR’s can be thought of as being derived from measurements taken in the hall. Because of all my processing and forcing mono IR's into a stereo pair, they are not scientifically accurate IR’s of the hall itself, but they do have the sound of a very good, very big concert hall. I’m sure there’s at least some resemblance to the original since I don’t adjust the original IR’s that much. In any case, you may find them musically useful if you want a large hall sound that is very friendly to orchestral music.

I’ve also used the set for surround reverb by using the SydneyStall1+4 for the front two speakers and the Choir 1+2 for the rears.

Not all the files on the source site are IR’s. To come up with my naming conventions, I downloaded only the IR-type files in the order they are in the table and sequentially numbered them noting which part of the hall they came from. The chart for understanding the seating locations is located in a PDF file called “Seating Plan” at this URL:

http://www.sydneyoperahouse.com/sections/about_the_house/venues/pdfs/seating_plan_concert_hall.pdf

Here is the code I used:

Stalls 1=J21, 2=P8, 3=Q26, 4=V37
Circle 1=B10, 2=H26, 3=M33, 4=S42, 5=T22
Boxes, 1=A23, 2=F26, 3=V26, 4=W26, 5=X25, 6=Y43
Choir 1=C35, 2=H15

For example, SydneyBox2+5 means that I used the IR’s from Box seats F26 and X25.

One final thought: If someone has the know-how and software necessary to convert the non-IR measurements on the site where I got the original IR’s to simple IR’s, we could potentially have lots more Sydney IR’s to play with. Help me out here! Further, I’m coming to realize that there could be a vast source of IR’s available if acoustic consultation firms in general could be persuaded to publish the data they collect on venues they work on before and after renovation. Acoustic firms might not be so reluctant to release the IR measurements (which we could convert to useable IR’s) if they didn’t have to divulge the name of the source room or building.