How do I make my own convolution impulses?

[Serenitynow]

I'm going to Italy, Montenegro, France and Spain in November. We'll be in some great acoustic spaces. I'm wondering how I can create some impulses of these acoustically cool places.

I know jack about this, but can I just clap my hands (or pop a balloon) and record it with my phone at it's highest quality setting, then later make an impulse out of it? Help!

[Swamp Ape]

I think most people use sine waves that move across the freq spectrum and record them with multiple mics in the space. I've never done it but have read some about it.

http://designingsound.org/2012/12/re...lse-responses/

Although I see here they use the clap/balloon method too.

[Serenitynow]

It looks like my basic clap plan may work ok.Then it says I need a deconvolver. It looks like the demo for the Voxengo could do it for me for free. Are there any other freeware VSTs out there? Does anything in REAPER do this?

I know there are a lot of impulses out there already, and I have several, but I thought it would be a cool, practical (and free) souvenir to have from our trip - 30th Anniversary!

[planetnine]

You can use an impulse!

-starter pistol or even two flat blocks of wood if it's quiet enough

An unfinitely short impulse is considered to have an infinite range of frequencies, so a sharp sound is a good source of all the ones we need.

The answer really depends on the software that uses the impulse. I've done them with wood smacks, but getting access and permission is usually the hardest part

-definitely a case of "who you know..."




>

[Serenitynow]

Originally Posted by SMM

Yes, "ReaVerb" (plugin in REAPER) will do this.

To find the Deconvolve feature you must:

+ Open ReaVerb.

+ Click the "Add" button in ReaVerb, then, select "File"

+ Click the "cancel" button in the pop up/search window.
.... After you click on the cancel button, You will see
the "Impulse Generation Utilities" appear.

+ Select the Deconvolve feature to "Deconvolve" your files.

Awesome! Thanks for the detailed info!

When I do the sound to record the impulse would I have the mic near my sound source or away from it?

[Serenitynow]

Originally Posted by SMM

The sound source and the mic placement effect the IR files.<<<<<<<<<<<<<<SNIP>>>>>>>>>>>
Side note (example):
I have read where people that were sampling the acoustic space of an Opera House,
placed the sound source on the stage, while the stereo mics (capturing the results)
were placed in the seating area (where the audience would be sitting).

That seems logical. Thanks for the link!

[joelsampson]

I would be inclined to buy a small field recorder to take. The results would be better than a phone and who knows what other great sounds you might want to capture. I have a little Sony I take with me. Runs a long time on a couple AA cells.

[whiteaxxxe]

... and get 2 packs of ENELOOP akkus (only from Sanyo) ... and a recharger for them. never ever buy any batteries no more.

[Serenitynow]

A friend from a guitar message board let me borrow a Zoom H2N which has 4 mics and can do XY, AB 4 channel surround etc. It can record in 24 bit 96khz. I'm excited about it now that I have a proper tool to use!

[sinkmusic]

Hi
The Zoom is a nice little recorder.
Just be careful with your levels and gain staging, as the recording level is set post A/D conversion, so any clipping is terrible.

[Judders]

Balloon popping works well.

Often, some editing is needed to the recorded pop before convolving.

[clepsydrae]

When you use balloons and starter pistols and such, you don't need to deconvolve. You're actually just creating the impulse file directly that way: the audio file is a recording of the instant spike of broadband noise (the pop sound) fading away. You just edit to the end of the spike of the noise, as Judders said. More info in the grey box at the end of this article: http://www.soundonsound.com/sos/jan1...notes-0112.htm

[edit for clarity: afaik you edit off the pop at the beginning, keeping just what comes immediately after until the end of the tail. (i've only done the sine wave method.)]

When you use the Voxengo program, it generates sine sweeps and then "deconvolves" those for you, but you need a loudspeaker with you to do that. Your method of recording with a handheld device probably lends itself to the simpler method so you don't have to tote a speaker around. But the results aren't as precise.

ReaVerb can deconvolve sine sweeps, but my loose information is that the voxengo program might do a better job.

And of course ReaVerb can use the results as convolution reaverb.

[hzandbits]

Those little Zoom recorders are alright - if a little heavy on the self-noise. But you might like a wider mic config than what those built in mics can provide. Consider getting a relatively cheap set of binaural mics (you know, earbuds with little mics in them). Typically they're omni mics, so if you can position them 2-3 feet apart, you get AB stereo. This will usually sound wider than X/Y (which is what the Zoom H2n has), especially in large spaces.
I actually built a pair myself from Primo EM172 electret capsules - very low noise, great sound, low cost. Here's a clip recorded with them: https://soundcloud.com/hzandbits/pos...t-2014-3-etape

[jazznfunk]

Can anyone explain how to pick a right test tone (sine sweep) lenght? ReaVerb's dafault is 4 seconds. Is there some connection between test tone lenght and final impulse response lenght? (small room vs. big cathedral)

[Xenakios]

Originally Posted by jazznfunk

Can anyone explain how to pick a right test tone (sine sweep) lenght? ReaVerb's dafault is 4 seconds. Is there some connection between test tone lenght and final impulse response lenght? (small room vs. big cathedral)

No, the tone sweep length is about getting a better impulse response. The longer the sweep, the cleaner the result can be, in theory. In practice using a longer sweep has the risk some unrelated noise like traffic or air conditioning will be included prominently in the sweep recording, making it worse for the deconvolution process(*). So it's a question of experimentation to find the ideal sweep length to use, really...

edit : (*) I may remember this wrong and it may be actually better if the sweep is longer if the recording space has lots of extra noise...There must be some article somewhere in the net about this...

[ashcat_lt]

Originally Posted by Xenakios

edit : (*) I may remember this wrong and it may be actually better if the sweep is longer if the recording space has lots of extra noise...There must be some article somewhere in the net about this...

Nah. I'm pretty sure that more noise in the recording just means more noise when you go to use the IR.

If you really want to get an idea of how these things work, I suggest using something more complex as your IR, like a drum loop or a short section of a full mix. Load it into ReaVerb and run something through it. I think you'll get the idea pretty quick. It's pretty much like triggering a sample of the IR based on the volume of the input, except that the sample is also sort of filtered by the input (and/or vice versa).

It can get pretty wild pretty fast, but it leads me to believe that you really want to avoid any unwanted noises in the initial recording of the IR.

[Xenakios]

Originally Posted by ashcat_lt

Nah. I'm pretty sure that more noise in the recording just means more noise when you go to use the IR.

This would certainly be the case if the IR is captured directly by popping a balloon, wood pieces or a gun...

However, I have a vague memory that with the sine sweep method the effect of undesired noise during the recording may be mitigated. (In any case, it would be a good idea to eliminate any extraneous noises as much as possible when doing the IR capture recording.)

[clepsydrae]

Originally Posted by Xenakios

However, I have a vague memory that with the sine sweep method the effect of undesired noise during the recording may be mitigated. (In any case, it would be a good idea to eliminate any extraneous noises as much as possible when doing the IR capture recording.)

This is my understanding/memory/expectation as well.

My impression is also that the (free) voxengo deconvolver is "better" than the Reaverb one, but i haven't done any head-to-heads. Has anyone?

I'm always confused about how the deconvolution algorithm knows when the test tone starts: meaning, if I record a space remotely, using a stock sine sweep .wav file that I generated with, say, the voxengo app, the deconvolution instructions never include any precise direction on how to align the left edge of the resulting room-response recording before runnin the deconvolution. Maybe there's 10ms before the sweep starts, maybe there's 2 seconds, depending on how I lined up the clip when i presed "record".

From that lack of instruction, I assume that it doesn't matter, and that the deconvolution algorithm can either figure it out or it doesn't even need to know. (I understand that once the IR is generated you can/should trim the head to suit your desires; i'm just talking about the deconvolution stage and wondering how it works -- i would have assumed that it would need to be precisely aligned with the sine sweep.) It all seems to work, so i guess it works :-), but i always have this nagging worry that i'm missing something.

[Serenitynow]

I'm the OP. I did take a few Impulse response captures while on our trip to Europe so I'm avidly listening to this thread as I will one day soon deconvolve and all that good stuff.

I only captured them with a loud clap so I'll be interested in how they turn out...when I finally have the time!

[insub]

I want to record the reverb IR of my church. I've been wanting to do this for a long time, but now that I've joined the choir I can get permission from the music director. And, I have an opportunity because this December I'll be taking my PA system in to broadcast the sound of the organ to the outside of the church.

There is no explanation of how to do this in either manual or the wiki.
However, I did find this article by SOS which describes precisely what to do using ReaVerb.
And, there's a help file that's accessible from within Voxengo Deconvolver which explains how to use that utility.

There were some comments in this thread indicating that Voxengo's Deconvolver may produce a better result. Any confirmation of such?

My speakers are JBL SF25 dual 15". Their frequency range is 35 Hz - 16 kHz (-10dB) and the frequency response is (±3 dB) from 45 Hz - 12 kHz. I intend to record the impulses using a pair of Behringer ECM8000 reference microphones. My plan is to have the mics at the typical listening position (center of congregation seating) and to record impulses from two positions, the alter (front) and the organ/choir loft (rear). You can see pictures on my church's site: http://www.stjameslou.org/ for an idea of what the inside is like. There is amazing and dense natural reverb inside. I plan to share the IR files here if/when the project is successful. I'm hoping to get a chance to do the same at the Cathedral of the Assumption but I believe that getting permission there will be more difficult.

Any advice beyond the SOS article above?

[karbomusic]

The recording part is easy so go ahead and get that done using claps or a sweep. If you are going to a church and don't have the ability to setup equipment to do a sweep, a hand clap will work fine. Getting the recording with a low enough noise floor is typically the biggest issue in public type places. Once you have the sampled sweeps/claps then you can spend whatever time needed to turn those into proper impulses.

[ashcat_lt]

Just understand that the sine sweep is going to give you an IR that represents that PA speaker at that place in that room - including frequency response, directionality, phase response, etc. That speaker is a bit limited in its response, and +/-3db can mean up to 6db between peaks in response and troughs, and that's on axis and probably only like a meter out.

Depending on what you're doing with it, it could be fine. Guitars might be pretty faithful, but if you run it on a drum kit, you'll probably hear the difference.

[insub]

Thanks!

Well the only speaker option I have is my studio monitors which are Alesis M1 MKII. Their response is listed as ±2 dB from 50 Hz - 20 kHz. My only concern with these are that I'm not sure if they will be loud enough.

The output is listed as:
Maximum Peak SPL per pair: ≥ 118dB SPL @ 1m
Maximum short term SPL: 80Hz to 3.0KHz ≥ 105dB SPL @ 1m

What do you think?

[ashcat_lt]

You probably don't want to push it all the way to the max output either. The louder the better of course for S/N purposes, but especially if the deconvolve process manages to take out some of the noise, you might get away with it. That speaker does have better HF extension, and actually gets pretty close on the low end, too. Might as well give it a shot. If you've got the time, maybe run both speakers and an impulse like a balloon or something. At worst you'll have a couple files files that you don't use, and at best you'll have options.

I left the microphone out of my thing above. The coloration of the mic will of course be imprinted on whatever you capture no matter what the source. A calibrated test mic meant for measurement would be best, but a decent flat condenser is probably good enough - closer to what you'd use to record the room in a live tracking session.

[clepsydrae]

I haven't done the clap method, I've only used the sweeps. I should try the claps, too... much more convenient. :-) AFAIK the big advantage of the sweeps is that they can filter out background noise like AC systems and such, which may be important depending on the background noise at the church.

I'd only suggest that once you have all the stuff set up and you're recording, might as well move the mics around a bunch and record a few different positions, and vary the separation between the mics and the angle, etc. Try X-Y vs ORTF vs. spaced pairs vs. crazy-wide-spaced pairs, try omni vs cardiod, etc. (oh, I guess your mics are just omni, so, just spaced pairs.) In my limited experience I couldn't easily apply my intuition about stereo recording to this application; meaning, the IRs I like the most are with mics in positions i probably wouldn't have used for stereo recording.

At any rate, if you're going to the trouble, it's nice to generate a few different versions of the space.

My current understanding and method, which i can neither support with reason nor provide links for :-), was cobbled from a bunch of random research on the net, and has come to:

- use voxengo deconvolver (if considering only free options)
- sine sweeps are the best for we mortals
- use 15 second sweep (longer is better for noise floor but the IR starts to smear due to "slight air pressure variations in the acoustic space. As a consequence, not only the IR is distorted, but also due to this smearing, the actual improvement of signal to noise is only by 2-3 dB for a 4 times longer sweep tone." -- link here.)
- use the "reversed technique" in deconvolver, and apply fade-in/fade-out in the UI of deconvolver when generating test tones

I think the main concern will be the speakers... and if the Alesis's can get loud enough, as you mentioned. You do want to use the flattest speakers you can find. You could always take both sets and try them both. Cathedral-like spaces are often more sensitive than people assume. If you can hear a reasonable verb level when someone is singing, you could argue that the speakers just need to match that volume, and it seems like they should be able to.

Another thought is that the ECM8000 isn't known for being low noise. If you have a nicer small-diaphragm condenser, that may be the way to go (or at least try), even though it's not a "measurement mic". You could also try longer sweeps if noise is an issue (if i have the physics right in my head.)

Excited to hear what you generate! It looks like a neat space:
http://www.heartofthehighlands.org/g...1786234423.jpg

I can imagine it sounds pretty great in there.

[karbomusic]

I haven't done the clap method, I've only used the sweeps. I should try the claps, too... much more convenient. :-)

It is. I've always used sweeps at home where I had the proper gear but I've also done some outside impulses such as a real mountain valley while on vacation and a sweep just wasn't going to happen. I found clapping and my Zoom or Sony PCM recorders worked just fine. It's reverb so unless we are being overly strict it really shouldn't be a problem. I've found noise floor to be far more the problem than frequency response YMMV.

[clepsydrae]

Originally Posted by karbomusic

I found clapping and my Zoom or Sony PCM recorders worked just fine. It's reverb so unless we are being overly strict it really shouldn't be a problem. I've found noise floor to be far more the problem than frequency response YMMV.

Yeah I have a zoom h4n now so i'm excited to start snagging some spaces while out and about. What zoom do you have? Have you had luck with the built-in mics for IR making? I have some nice little super-portable electret-to-phantom/XLR mics i made that should work well... just curious about the preamps in the zoom for this application.

I recently put my monitor up against a piano with the sustain pedal down and did a sweep. Deconvolved the recording, cut the head off the resulting IR, normalized the remainder. Voila: spacey piano reverb. :-)

[karbomusic]

Originally Posted by clepsydrae

Yeah I have a zoom h4n now so i'm excited to start snagging some spaces while out and about. What zoom do you have? Have you had luck with the built-in mics for IR making? I have some nice little super-portable electret-to-phantom/XLR mics i made that should work well... just curious about the preamps in the zoom for this application.

H4N as well. As of late I've been taking it to band rehearsal because it has phantom power and I plug in my V67Q (stereo) LDC into it to record. Works great!

I recently put my monitor up against a piano with the sustain pedal down and did a sweep. Deconvolved the recording, cut the head off the resulting IR, normalized the remainder. Voila: spacey piano reverb. :-)

Nice! I'd need to do some serious digging but somewhere around here I have a recording where I took a piezo and attached it to a cymbal essentially making the cymbal into a deranged microphone.

[clepsydrae]

Originally Posted by karbomusic

H4N as well. As of late I've been taking it to band rehearsal because it has phantom power and I plug in my V67Q (stereo) LDC into it to record. Works great!

Nice! Yeah I'm kinda enamored with my h4n. That 4-capsule mic i made over in the DIY thread was designed to optionally plug into it: 2 channels into the 1/8", 2 into the XLR.

[insub]

Thanks, all.

Unfortunately, the ECM8000s and other above listed items are the best I have access to.

The space is really awesome.
I was impressed when it was just singing and the organ, but the day we had a professional trumpet and violin player join us in the loft I was simply blown away! I wanted to get the IR before, but from that moment on I knew I had to make this happen one way or another.

[insub]

I've read that most of the problems people had with the handheld recorders are handling noise. Do you use some type of stand and shock-mount?

[clepsydrae]

Originally Posted by insub

Unfortunately, the ECM8000s and other above listed items are the best I have access to.

No problem; go for it. I'm sure it'll work out. Or at least you'll learn something in the process. :-)

And musiciansfriend.com has (or at least had) a very generous return policy. :-)

Originally Posted by insub

I've read that most of the problems people had with the handheld recorders are handling noise. Do you use some type of stand and shock-mount?

They often come with adapters to mount on stands, so you can do that. (Don't know that you'd need a shock mount, though.) Or just set them on a table or wedge them in a tree branch or whatever.

[karbomusic]

Originally Posted by clepsydrae

Nice! Yeah I'm kinda enamored with my h4n. That 4-capsule mic i made over in the DIY thread was designed to optionally plug into it: 2 channels into the 1/8", 2 into the XLR.

Ah right!

[insub]

In Voxengo's instructions it says that the test tone file and the impulse file must be the same length overall accounting for presumed reverb tail length.

How do you handle this?

Do you just use "Glue items within time selection" to make the test file as long as the recorded file prior to deconvolving?

[karbomusic]

Originally Posted by insub

I've read that most of the problems people had with the handheld recorders are handling noise. Do you use some type of stand and shock-mount?

I just put it on my camera tripod or mic stand IIRC.

[clepsydrae]

Originally Posted by insub

In Voxengo's instructions it says that the test tone file and the impulse file must be the same length overall accounting for presumed reverb tail length.

How do you handle this?

Do you just use "Glue items within time selection" to make the test file as long as the recorded file prior to deconvolving?

Speaking from memory, i believe you have misinterpreted (unless i have misinterpreted what you're saying): "For deconvolution to work, you should record the full test tone duration without any cutouts. For reverberant impulse sources, you should record additional silence which should be at least as long as the expected reverb tail."

In other words, the recording will be longer than the test tone, so, not to worry; no need to make them the same length. (But point me at the instructions you're talking about if that doesn't seem to line up.)

Weirdly, you also don't need to clip the front of the recording to some precise place: when generating the file that you're going to deconvolve, just give a couple/few hundreds of ms before the recording starts showing signal in the waveform, and as much tail as is needed to capture the verb and then a bit of a buffer. Somehow the deconvolver sorts it all out. You can then edit the IR file (which is just a regular audio WAV file, remember) to remove silence at the head and, if you desire, fade out the reverb tail to whatever length.

On the subject of recording length I did see this which was interesting: "Make sure you record enough silence after the test tone ends. Ideally, the duration of this trailing silence should be 1.5 to 2 times more the expected reverb tail's length. For a pre-amp or other hardware unit you may additionally record at least 1 second silence." For, say, a 4 second reverb, I haven't been leaving 8 seconds of silence after the tail! Not sure it would make an appreciable difference, though, really.

[clepsydrae]

Originally Posted by clepsydrae

You can then edit the IR file (which is just a regular audio WAV file, remember) to remove silence at the head

Putting a finer point on this: i find this method (leaving a tiny bit of buffer silence at the start of the edited recording to be deconvolved, and then editing the IR) to be easier than precisely clipping to the start of the tone in the recording file to be deconvolved, but ymmv.

[insub]

Thanks for clearing that up. I had thought it was talking about the test tone needing silence for the tail, which did not make sense. And, I had not noticed that it said the silence needed to be at least as long as the tail. I'll be sure to print out the instructions when I go.

Thanks for the tips!

I'll probably try a clap & balloon pop while I'm at it so I can compare the resulting IRs.

[clepsydrae]

Originally Posted by insub

And, I had not noticed that it said the silence needed to be at least as long as the tail.

Yeah, i'm not sure that's really a "need" so much as an ideal... if anyone knows about that, please chime in. It would seem to me that covering the tail plus a second or two would work for any reverb situation. (Certainly has for me so far.) They make that comment in a FAQ about the deconvolver sometimes outputting empty files because there wasn't enough silence at the end. I wonder if it only really happens with very short reverbs or device modeling. It's hard to imagine there would be any use for 25 seconds of silence after a 10 second tail has died out, when recording e.g. a cathedral. You could of course always just tag on some silence in the DAW if you had to, so the whole thing seems a little weird to me, unless there's some esoteric value in the quasi-silence of the room sound for noise issues or whatever.

The reason i belabor it is that when doing lots of IR recordings it's nice to be able to do them quickly, and nice not to have to worry about capturing "silence" of ambiguous definition, especially when a dog might bark or someone walks in or whatever. The shorter the sweeps, the easier everything gets.

I'll probably try a clap & balloon pop while I'm at it so I can compare the resulting IRs.

Yes, and please share the results so we can all learn. :-)

[ashcat_lt]

Reverb never actually dies out, it just gets too quiet to tell from noise. If you stop recording at that point, and the deconvolution removes some of that noise, then it just might end up as an abrupt cutoff. Might not be an issue in most situations, but if maybe you want to do something crazy like smash the reverb into heavy limiting/distortion? You'll hear it then most likely.

In fact, this is one of those situations where I would recommend trying your best to "use all the bits" if at all possible. Record the impulse so that the loudest peak is as close to 0dbfs as you can get without adding much distortion. Maximize your dynamic range in the capture to give you the most options down the road.