Techniques for Recording Impulse Responses of Real
Acoustic Spaces
by David Gauger II
"Shofarsogood"
(See the previous article for microphone selection, placement and
geometries when recording impulse responses.)
Impulse Generation
Now that your mics are set up and the lines are run, there are 3 main
methods to generate an impulse that will garner the response we're looking
for. The oldest is a simple gunshot. There are several problems with this
method. You need to carry what looks like a firearm (not always a great
idea) which has gun powder residue all over it (a problem if you fly a
commercial flight!) and is comparetively expensive. And on top of it all,
the impulse it makes (the gunshot) is not always consistant and might not
contain every frequency at equal energy which is what we're after. In
addition, not every gunshot has the same exact intensity. On the plus
side, it's loud (we need lots of volume in order to get a decent signal to
noise ratio), it's small, and the resulting digital recording IS the IR:
no post processing necessary other than to edit the beginning and end of
the IR.
Another method is called the swept sine. The basic idea is to take the
frequency spectrum and spread it out over time. The result is a sine wave
that starts at 20 Hz and sweeps up at a predetermined rate to 20 Khz. On
the plus side: if you've got a good playback system, this method yields
extremely accurate spectacular sounding IR's. The downsides: much more
equipment to set up (amps, monitors, CD player, etc.) and you must have
accurate monitors capable of filling whatever room you are impulsing with
a lot of sound. The pros use omni-directional speakers to send out energy
in all directions. Also, after you're done recording, the sweep tones must
be run through a piece of software called a deconvolver. Audioease
Altiverb has one, as does Voxengo and a few others. The output of the
deconvolver then often needs to be edited (heads and tails) also.
The third method, MLS sequence, uses a specific sequence of pulses that
are then run through a deconvolution-like process to recover the IR. This
method is least used by amateurs due the the software needed to utilize
it.
Impulsing tool (substitute for a starter pistol)
First try: starter pistol. They’re not readily available these days.
Besides, showing up at some church and shooting what looks like a real gun
(even if it’s a starter pistol) with all the noise and smell of spent
gunpowder is enough to make people really nervous these days. They also
cost about $100, at least as far some cursory web research went.
Bottom-feeder solution: a powder actuated power nail device. The
Remington company makes one for about $30 and it’s designed to drive nails
into concrete. Basically, you put a .22 blank in the top, a special
case-hardened nail in the bottom, then tap the top of the device with a
standard hammer. The charge blows (making the big bang that we’re looking
for) driving the nail into concrete.
The .22 blanks come in strengths numbering from 1 to 4. Use the #4
charge which is the strongest. Strength number 3's didn’t sound very good
to my ear. The problem with #4 charges is that they really drive the nail
hard. One solution is to fire the nails into a double thickness (about 7
or 8 inches thick) of a 2 foot long piece of “Glue-Lam” board. All the
laminations and glue holding them all together slows the nail down better
than just a 6 by 6 board.
Problems notwithstanding, we now have a legal, unobtrusive device that
makes a lot of noise similar to a starter pistol. The tool, .22 blanks,
and nails can all by purchased at your local ACE or Sears hardware stores.
Here is the device: http://www.cornerhardware.com/productdisplay.pl?item_id=215436&AID=10283904&PID=1255551
One issue here is the tone quality of the resulting “gunshot”. The
shots are fairly consistent although there are some that could be
considered “duds” or at least they lack power and sound different than
normal. Do at least 2 “gunshots” at each impulse location to account for
these variances. In the editing stage, choose the gunshot that sounds most
like all the others in the set in terms of tone and volume.
As for the frequency content of the sound, they’re a little bass shy
and perhaps a tad mid-range heavy but still very useable. Perhaps an RIAA-like
emphasis EQ curve could be designed and used on each impulse in the
editing stage. The frequency content mimics the fact that many engineers
roll off the bottom end of the feed to the reverb anyway to avoid bass
buildup.
With all the talk about Gigapulse's ability to add reverb from
different originating locations, you can blanket the stage doing a nailer
“gunshot” every couple of feet or so. The idea is that if you want to have
a sound that is panned hard left, you can use an IR where the gunshot was
recorded from a hard left position as well, and you'll get the hard left
early reflections and coloration. That is the way it occurs in real life,
as we all know.
Recording Equipment
Use whatever you have, as long as it's digital because the signal to
noise ratio is not favorable to analog equipment noisiness at the end of
the reverb tail. You can even use a computer if you have enough mic pres
and can quiet the fan noise somehow. The Fostex VF16 16 track digital
recorder works well for a number of reasons:
Disadvantages: very poor internal effects (my opinion) and fixed 16 bit
44.1 kHz recording.
The VF16 has 8 inputs but only 2 of them have XLR connectors. This
means that it lacks mic pre’s and and phantom power on the remaining 6
inputs. I use the SM Pro Audio PR8 unit to provide this. It is one rack
unit high and has 8 mic pre’s with switch able phantom power on each
channel. Not the quietest mic pre’s on the planet, but OK if you make sure
you've got your gain staging right.
To record on the remaining 8 channels, the VF16 requires you to input
the sound via the ADAT port. The Behringer ADA8000 does the job here. It's
a one rack unit A/D and D/A converter with phantom power, 8 XLR’s, 8 mic
pre’s, balanced outputs and ADAT output as well. All for $199. What a
deal. The mic pre’s sound reasonably good, too. Bottom line with this
setup: portable 16 track digital recording, and I don’t have to rip the
computer out of the studio to record.
As far as setting levels goes, the typical rules apply. Record as hot
as possible without clipping to maximize signal to noise ratio. It goes
without saying that you'll probably want to record at the highest sample
rate and bit depth your equipment will allow.
Recently I've acquired equipment that enables me to record at 96 Khz,
24 bits and also some Mackie HR824 monitors. This jump in equipment has
made it possible to record 45 second sweep tones and recover the IR files
as high as 96 kHz, 32 bits after editing.
Editing
In general,edit your "gunshot" IR's so as to leave the spike in. Some
say this is good since you get localization and can use the IR's fully
wet. Others disagree. The truth of the matter is that it's just plain
easier to leave it in. Trying to find where the impulse (spike) ends and
where the response (reverb tail) begins is not at all obvious. My advice:
leave it in. You'll find the sound is not compromised that much, if
at all.
Also, since you’re probably running more than one mic pair, if you
preserve the timing between the impulse pairs, you can use two pairs of
impulses in a 4-channel surround situation. I do it by starting with the
nearest mic pair. If the location of the beginning of the impulse is at
1:03:04.332, then I edit ALL the impulse pairs using that as a start
location. As you move back in the auditorium, the sound arrives later and
later to the far mic pairs with respect to the nearest pair. The result is
that there is increasing “dead air” at the beginning of the far mic’s IR’s
but if you edit even the far IR’s with a start time of 1:03:04.332 and
leave the dead air, all your IR’s will be time-aligned.
I generally try to leave at least one second of dead air in the file
after the last of the sound has died away in the IR for the convolver and
it’s latency.
Summary
To sum all this up:
1. Use matched condenser mics.
2. Choose a mic geometry. Spaced omnis and rear-facing spaced cardioids
work well.
3. Choose mic locations between 1/3 and 3/4 or more of the way back in
the hall.
4. Measure and document everything. Pictures are good.
5. Set your levels, hit record, then blast away. Blanket the stage with
impulses.