posted 21 October 2001 23:56            

I just got an Earthworks M-30 measurement Mic and I tried some experiments to obtain the room response. One of these involved popping a balloon and recording the "impulse response" of the room.

I notice that the initial pulse is bipolar -- meaning that it is a rising pulse followed immediately by a negative going pulse at about the same amplitude, and then followed by some ringing before the first few reflections start arriving. I suspect that this has to do with (a) the finite duration of the impulse -- a balloon popping is not a perfect impulse, and (b) conservation of mass -- which would imply that the sum amplitude through a pulse of air would have to equal zero.

Anyway, when I try to convolve music using this measured impulse I end up with a very annoying high-pass filtered sound, not at all like music played through loudspeakers in this room. The bipolar pulse at the front of the recording would indeed have a high-pass character to it.

If I convolve white noise with this pulse then the resulting noise is very colored and displays peaks at frequencies where I would expect the room modes.

It appears to me that I don't really have the "impulse response" but rather the convolution of the balloon popping waveform and the true impulse response of the room. I can guestimate the location of the first few reflections and create an artificial impulse response file containing those first few impulses. When I do that the results sound a bit more like the actual room -- very comb filtered.

So my questions to all you audio experts out there are these...

When you want to get a sample of live reverb, do you use a popping balloon, or some other technique?

And if you use a popping sound, do you do any kind of preprocessing to the recorded reverb (like I did with eyeball estimation of reflection impulses) or do you actually use the raw recording itself?

And if you use the actual recorded waveform for your reverb convolution kernel, how do you deal with the high-pass character of the sound?

Cheers,

posted 22 October 2001 07:23            

We also use pink noise to get the frequency responce correct, and check to see that the clicks are telling us some sort of truth.

A balloon is quite a bassey bang and will make a not so instant sound even without reflections. A hand clap would be better.

Obviosely if you use a (44.1k) one sample wide pulse from a normal D to A converter, it would be so affected by the output filter, and attenuated so much, that it would probabley be unuseable.

posted 22 October 2001 18:21            

I think this is the first time, Iīll be able to tell you something

Most impuls responses are mesured using the MLS method. devices like the MLSSA play a pseudo random noise that results from a modulus-2 division. it has a certain length (lets say 256 bit) and loops. this is played through a mesuring speaker and sampled through a mic. the software "subtracts" the signal from the pseudo-noise it knows. the result is the impuls response. i donīt have the background to explain to you how this is done but the advantage is that you donīt have to generate the perfect dirac-impuls (infinitely short while infinitely loud) you need to mesure the impuls response plus it is almost imune to noise that comes from the room itself (people talking, rumble etc.)
iīve used the machine quite a while ago and it is easy because you can compute all other parameters like the RT60 etc. off-line from the impuls response.

i hope this helps all though not very scientific

posted 22 October 2001 19:55            

I have a few comments here...

1. The balloon popping experiment produced an excess of treble and nearly -24 dB bass response relative to the high end. Very surprising result, and not what I might have expected. But the leading pulse is clearly a high-pass bipolar spike.

2. I later did try sending a Kyma impulse to the speakers, and the results recorded acutally worked out pretty well -- a lot of bass response (maybe too much) but the impulses sound like this room when convolved directly with musical material. I realize this measures both the speakers and the room together...

3. I use something called SIA Smaart which is very much like MLSSA for making room measurements. But in this case I didn't really want the Fourier Transform of that response, but rather the impulse response in the time domain so that I could use it for convolution in another program that needs the time domain responses. Actually for use with Kyma, the frequency domain response would be more appropriate!

You are correct about the use of maximal length psueudo-random pink noise for testing room responses. It is easy to show that the cross correlation between the source sound and the recoreded sound is the transfer function of the room. The math is pretty straightforward on that one.

But I saw a quick video a while ago on a new Sony hyper-realistic reverb unit they just produced, and it uses impulse response data recorded on CDROM's. They have real responses from cathedrals and concert halls from all around the world, and they even recorded in a special place in the Grand Canyon. I thought I noticed them using chirps for recording the response, and maybe some balloons filled with helium and tied around various places. But my memory may be faulty.

Anyway, I find that the balloons probably have too slow a popping behavior to produce really good impulses. I'll give the handclap idea a good try here too.

I should know how to use chirps for this purpose, based on all my ancient radar processing knowledge, but I have to refresh that some more here. I did notice a good cross correlation last night between the source chirp from Kyma and that recorded by the M30 across the room. I was able to see an 11 ms delay, but the fine structure in the reverb and echoes alludes me for the moment. I can see them clearly in a sonogram display, and the reverb tail looks like it lasts about 100 ms, but I can't get the details about the reflections just yet.

As it happens, my November issue of Electronic Musicial mag just arrived and they feature an article talking about room treatment using parametric EQ's. I'll have to finish that article before I know how they test the rooms electronically.

Thanks guys! Always looking to learn a trick or two from you pro's!!

Cheers,

posted 23 October 2001 14:02            

There is an interesting Web site of a division of Lexicon, known as Lares, and headed by a Dr. David Griesinger, that specializes in measuring performance halls and providing active reverberation correction.
http://www.lares-lexicon.com/

A very interesting paper appears in the collection of scientific articles, detailing the process of making hall measurements, "Beyond MLS - Occupied Hall Measurements with FFT Techniques", Griesinger, 1994.
http://www.lares-lexicon.com/introtest.html

That paper discusses balloons, handclaps, and pistol shots as potential impulse sources. He concludes that the best approach is a swept sine source that must be deconvolved for useful information.

Most notably, that paper points out the difficulties in making measurements below 300 Hz, with all of these impulsive sources lacking substantial enough energy at these low frequencies. That pretty much confirms what I found here with the balloons -- no low frequency energy. So my results were tainted by lack of spectral excitation at the low frequencies...

Reading between the lines, I get the impression that he is or was a Professor at MIT and that he is one of the founding members of Lexicon Corp. in Waltham, MA.

Cheers!

- DM

posted 23 October 2001 23:13            

YES!! I would love to read his thesis! 5.1 convolutions sound like real killers to me. I would be very interested to read how this is done.

Thanks!

posted 24 October 2001 05:35            

itīs not *really* realtime. it needs about 1-2 seconds to compute the convolution.

posted 24 October 2001 11:37            

But thanks! I went to their Web site and downloaded their Huron manual to see what it is all about.

Cheers!

-- Yes, indeed! I remember these guys, and I remember feeling irritated with them as I do now -- about their claims of "proprietary Lake convolution algorithms". Hey! Convolution is convolution, and it's been around for at least 200 years. What's new since about 1960 is fast convolution using FFT's for block overlap/add or overlap/replace convolutions of large data sets.

But all in all, it looks like convolution is their main area, and they offer a mainframe that seems very similar to Capy. However, it appears that the programming is mostly DSP Assembly code. It also looks VERY expensive -- maybe someday I can afford one in my dreams!

- DM

PS: It looks to me like a Capy could run circles around their Huron system. They use 56002 DSP's and have a lower limit of available DSP power than a Capy. The only major difference appears to be their programming tools -- DSP Assembly with Win/NT C++ capability. HINT, HINT, ... FLI (Foreign Language Interface) would be really keen for Capy!

posted 19 November 2002 11:05            

i believe they are using a track and field gun shot for their ir sampling