Impulse Responses: Normalize vs. Let 'em Alone

[osiris]

I take that back. I did hit the Normalize button once. I just didn't see or notice any difference in the sound. TBF it was a short verb on a snare.

[DaveClark]

With IR captures made with microphones in real spaces, you “excite” the space with some kind of brief loud noise burst. Like playback through a speaker, a burst balloon, hand clap, etc.

Since 2000, many if not most people use the Exponential Sine Sweep method which involves a swept-frequency sinusoidal wave and deconvolution pioneered by Angelo Farina:

A. Farina, “Simultaneous Measurement of Impulse Response and Distortion with a Swept-Sine Technique,” presented at the 108th Convention of the Audio Engineering Society (2000 Feb.), convention paper 5093.

http://pcfarina.eng.unipr.it/Public/Pap ... -AES00.PDF

There are test signals, MATLAB code, and all sorts of stuff available for using this technique. It is particularly useful for "mildly nonlinear" systems because you can separate out the nonlinear stuff.

[MTorn]

I’m fairly ignorant when it comes to the sine sweep method. Does it produce a different sounding WAV file? If the audio for those have an audible sine sweep, then I’ve never come across one while assembling my collection.

If you want to use a sine sweep to collect responses from real life spaces, I imagine it requires a pretty flat response from the loudspeaker playing the sweep.

[waiting man]

I’m fairly ignorant when it comes to the sine sweep method. Does it produce a different sounding WAV file? If the audio for those have an audible sine sweep, then I’ve never come across one while assembling my collection.

If you want to use a sine sweep to collect responses from real life spaces, I imagine it requires a pretty flat response from the loudspeaker playing the sweep.

Slight risk of all kinds of artefacts might creep in, yes.
Even the sine playback must be pristine, D/A, output levels, whatever.
All numbers must check, in theory.

Imo, you'll always need all kinds of tone (lp and hp practically mandatory) or pan or predelay control with those ir things, so go with the flow I say.

[DaveClark]

I’m fairly ignorant when it comes to the sine sweep method. Does it produce a different sounding WAV file? If the audio for those have an audible sine sweep, then I’ve never come across one while assembling my collection.

If you want to use a sine sweep to collect responses from real life spaces, I imagine it requires a pretty flat response from the loudspeaker playing the sweep.

The method produces IRs that have the same type of shape as those obtained by other methods. This is due to the deconvolution step which calculates an impulse response from the source signal (sweep) and response (altered sweep).

All methods produce IRs that represent the entire system that is being measured, which includes more than the specific component of interest (room, cab, ...). Even popping a balloon is a complex process that starts with pinprick and ends up with some type of gas going all over the place while the balloon shrinks down, flying this way and that while sputtering. It's not a Dirac delta function.

[cthonophonic]

Also, the spectrum of a popped balloon or (especially) a clap will often be very colored, and the sound energy is not always directed uniformly, so there will be slight differences in the stereo field of the recording. I've found that you can average that out to some extent by stacking multiple recordings and carefully phase-aligning them, but a sine sweep is consistently flat (or as flat as a given speaker/mic configuration will allow).

In my experience trying the sine sweep method outdoors, the main thing I've discovered is that coyotes absolutely love it. One day, I will get a single clean sweep recording, and it will be glorious.

[MTorn]

Aha, now I understand the finer details between using noise bursts and a sine sweep. There's no way of telling the difference by listening to the IR files (once the sine sweep version has been deconvolved), other than that the results might be of higher quality. I'm guessing that the sweep method helps remove things like coughs and squeaks?

Still, in the cases where a noise burst remains at the onset of a reverb IR WAV (a per peeve of mine), I'm assuming that those used the non-sweep method, and I usually crop the burst off.

This far I've only made IRs from electronic and virtual gear (with a noise burst), but now I'm itching to grab some acoustic spaces.

About the original question of normalizing - this was probably already mentioned, but I can only think of one reason I wouldn't do it: If you have a collection of IRs that have already been level matched (by ear, or some other method better than brute force normalization), and want to preserve their relative volume.

[DaveClark]

I'm guessing that the sweep method helps remove things like coughs and squeaks?

Still, in the cases where a noise burst remains at the onset of a reverb IR WAV (a per peeve of mine), I'm assuming that those used the non-sweep method, and I usually crop the burst off.

The following article has sections in which Angelo Farina discusses how to handle pre-ringing (Section 5.1) and pulsive noises during recordings (Section 5.3) when using the exponential sine-sweep (ESS) method.

http://pcfarina.eng.unipr.it/Public/Pap ... nd2007.pdf

I don't myself bother with this kind of exacting detail, not being in the business of selling professional IRs as such. I'll wait for the day when some listener tells me that they couldn't sleep the previous night because a snare in one of my pieces didn't have the right transient sound due to an IR I used, and they can tell me exactly why and what to do about it.