Convolution?

Reaktor, MAX/MSP, VST/AU, etc. A place for all things soft....

Moderators: Kent, Joe., luketeaford, lisa

User avatar
fisherking111
Veteran Wiggler
Posts: 583
Joined: Mon Jun 25, 2012 2:45 pm

Post by fisherking111 » Mon May 05, 2014 12:17 am

Here's a link to a jam I just recorded using using this IR I just made from the Quadraverb:

https://www.dropbox.com/s/wmujb1tkigqa3 ... lution.mp3

M4L LFO modulating the Width, pre-delay and panning of the convolution reverb. I made 4 different IRs with slightly different volumes, which I switch out throughout the recording. Also, I change the Size and Decay of the convolution effect throughout.

So I have the TX802 arpeggio going straight to the convolution via send, and another channel (the more in your face, not panning part) going through the Quadraverb live (you can hear the major tonal difference, even with the modulation). Weird feedback at the beginning from a recording feedback loop, final 10 seconds is just the TX802 samples ITB, no quardraverb or convolution (you can hear how vanilla it sounds compared to the the punchiness the Live Quardraverb gave it). Most of the track the Quardraverb is on bypass, with the input slightly in the red.

@3:40 you can hear the convolution bit by itself, then the quadraverb channel fades back in.
[bandcamp width=350 height=470 album=3759068727 size=large bgcol=333333 linkcol=fe7eaf tracklist=false]

User avatar
Ginko
Super Deluxe Wiggler
Posts: 1579
Joined: Thu Jul 18, 2013 6:27 am
Location: Unfortunate
Contact:

Post by Ginko » Mon May 05, 2014 1:22 am

Just to mention it again, you can try a free version of Valhalla DSP vintage reverb - what you are doing sounds fun though :D

VibratingMotorGate
Common Wiggler
Posts: 74
Joined: Tue Jul 16, 2013 9:30 pm
Location: Berkeley, CA

Post by VibratingMotorGate » Mon May 05, 2014 4:21 am

wouldn't a convolution of the converter stage in something like a classic hardware reverb or yamaha FM synth be pointless for achieving an even remotely similar sound though? Those architectures rely on feedback loops and intermodulation from these non-linearly error-ridden circuits which doesn't just simply change the the tone, but the fundamental and is harmonics completely shift and change in the time domain. A convolution isn't going to capture any of what makes those sounds special.

User avatar
subbasshead
Veteran Wiggler
Posts: 745
Joined: Sun Nov 16, 2008 3:49 pm
Location: Wellingtron, New Zealand

Post by subbasshead » Mon May 05, 2014 4:34 am

Anyone here ever make an IR not related to reverb tho?

For film sound design I've experimented lots with convolution

eg for a horror film I made an IR from a loop group/ADR recording of a group of Japanese girls whispering, and then convolved it with wind sounds = creepy as fck wind with whispery elements!

The thing with convolution to remember is convolution reinforces common frequencies so eg convolve a sub rumble with a violin & you get nothing..
but convolve two sounds with similar frequency densities & you get very interesting results...


While the idea of capturing reverbs & gear etc is very useful, putting all sorts of things into IRs is very good for experimentation....

As a learning experience its also very valuable to just listen to the raw IRs... eg find one of the beautiful TC or Lexicon verb IR libraries online & then audition the raw IR .wav files - its a bit like minimalist dubby one shot raster noton... ie a single ping delayed/verb'd

Also there is a great IR collection you can buy by Spirit Canyon called Sanitarium - the original site seems to have disappeared but this thread:
http://www.kvraudio.com/forum/viewtopic.php?t=75480

their IRs are very useful, for making nasty tonal elements.... but listening to the raw IRs, most have a lot of 'noise' ie broad spectrum, so no matter what you throw at them there is always some common frequencies....

serious fun!

User avatar
Carl Hungus
Veteran Wiggler
Posts: 613
Joined: Mon Oct 13, 2008 1:57 pm
Location: Logjammin'.....

Post by Carl Hungus » Tue May 06, 2014 6:30 am

I wouldn't use Speakerphone for this, rather Altiverb. Same company, but you can make IRs more easily I think with Altiverb than Speakerphone, and it's frankly more well suited.
I have both, but I find Speaker phone to be great with post production loop grouping, or futzing phone conversations etc; generally post production use. I do use it for creative mixing, but the price is a bit high for the amount I use it in music. Altiverb I use on everything.

Also Altiverb ships with IRs of FX units and spring verbs, plus a bunch of other "creative" impulses like inside pianos, or toy echo boxes.

I think making the impulse is easy, just run their impulse file, record that and their software removes the frequencies of that file and you're left with the impulse.
It can be a piece of gear or a mic'ed room, doesn't really matter.

User avatar
madronalabs
Common Wiggler
Posts: 210
Joined: Sat Mar 19, 2011 1:31 pm
Location: Seattle
Contact:

Post by madronalabs » Wed May 07, 2014 12:48 pm

As a learning experience its also very valuable to just listen to the raw IRs... eg find one of the beautiful TC or Lexicon verb IR libraries online & then audition the raw IR .wav files
I second this! Listen and look and learn what makes the nice sounds nice.

I have some EMT 250 IRs that I use a lot on vocals. Even though they don't give the liveness of the real thing, they have a different more static sound that I like a lot.

User avatar
leeski
Super Deluxe Wiggler
Posts: 2124
Joined: Sat May 12, 2012 5:44 pm
Location: Saturn

Post by leeski » Wed May 07, 2014 1:29 pm

madronalabs wrote:
As a learning experience its also very valuable to just listen to the raw IRs... eg find one of the beautiful TC or Lexicon verb IR libraries online & then audition the raw IR .wav files
I second this! Listen and look and learn what makes the nice sounds nice.

I have some EMT 250 IRs that I use a lot on vocals. Even though they don't give the liveness of the real thing, they have a different more static sound that I like a lot.
I always goto the EMT verbs in Altiverb they sound right all the time.

User avatar
fisherking111
Veteran Wiggler
Posts: 583
Joined: Mon Jun 25, 2012 2:45 pm

Post by fisherking111 » Thu May 08, 2014 1:36 am

I was hoping for a "digital shortcut" to get the A/D D/A sound of my quadraverb rack... but I think there is no shortcut...

Ironically, I can see 5-10 years down the road, some company making "super realistic" DSP hungry plugins to model old device inputs and outputs...

It seems like Roland's TR-8 and TR-3 are step in this direction, if they are truly modeling the entire circuit flow in code as I've read they are...


Lots of more experiments to do with IRs and Convolution though! I love the sound of these reverbs! I downloaded about 1GB of free ones from different sites I'm still going through!

I recommend checking out freesound.org for some interesting real space IRs people from around the world have uploaded to share!
[bandcamp width=350 height=470 album=3759068727 size=large bgcol=333333 linkcol=fe7eaf tracklist=false]

User avatar
secretkillerofnames
Veteran Wiggler
Posts: 593
Joined: Wed Apr 03, 2013 12:07 am
Location: Brisvagueness

Post by secretkillerofnames » Thu May 08, 2014 4:49 pm

fisherking111 wrote:I was hoping for a "digital shortcut" to get the A/D D/A sound of my quadraverb rack... but I think there is no shortcut...
From what you are saying it sounds like you should check out Nebula... however there is no cheap / easy entry point to that and the results are so subtle you may well wonder if you've been scammed ;)

However I just did a remix and used the STN Nagra, some of the Henry O chains and some of the Cupwise Tube Radio impulses and I can definately hear the impact. Mix sounds more spacious, there is a lot more movement in the upper mids.
I recommend checking out freesound.org for some interesting real space IRs people from around the world have uploaded to share!
The OpenAir library has some quirky spaces as well.

Has anyone purchased the Diego Stucco Convolution Processing impulses? I'm keen to get these on pay-day and run some material through them in Reverberate.

User avatar
secretkillerofnames
Veteran Wiggler
Posts: 593
Joined: Wed Apr 03, 2013 12:07 am
Location: Brisvagueness

Post by secretkillerofnames » Thu May 08, 2014 4:58 pm

fisherking111 wrote:I was hoping for a "digital shortcut" to get the A/D D/A sound of my quadraverb rack... but I think there is no shortcut...
Also I notice there are some Quadraverb presets in the Nebula library - all for Modulation and Reverb. What you probably want is the "Mojo" something Henry O does well but don't think he's done it with a QV

kaariainen
1-Post Wiggler
Posts: 1
Joined: Thu May 08, 2014 5:10 pm
Location: Finland

Post by kaariainen » Thu May 08, 2014 5:18 pm

Thanks for the links!

User avatar
tilman
Common Wiggler
Posts: 109
Joined: Thu Oct 06, 2011 8:56 am
Location: Germany / Cologne

Post by tilman » Wed May 14, 2014 4:16 pm

secretkillerofnames wrote: Has anyone purchased the Diego Stucco Convolution Processing impulses? I'm keen to get these on pay-day and run some material through them in Reverberate.
Bought them today. Not expensive at all... listening the files teaches you what is important for creating experimental convolution files... But on the other hand it is so simple to create your own convolution snippets.

One cool thing in Kyma there is a module that allows you to do realtime convolution. The convolution signal is read in realtime. So you can convolve two live audio streams.....

User avatar
lessavyfav
Ultra Wiggler
Posts: 990
Joined: Wed Jun 02, 2010 11:00 pm
Location: Brooklyn, NY

Post by lessavyfav » Wed May 14, 2014 4:29 pm

Hat Diego Stucco stuff is cool. On a bender I bought his 2 tutorials in the subject of creative convolution. I don't recommend them though. They are nicely done and interesting but rather shallow - 9 buck for a 14 min walk through without source files just felt a little light. I didn't get the impulses since DIY seems fun and easy
I'm Tim and I'm excited for modular synthesis! :nana:

User avatar
secretkillerofnames
Veteran Wiggler
Posts: 593
Joined: Wed Apr 03, 2013 12:07 am
Location: Brisvagueness

Post by secretkillerofnames » Wed May 14, 2014 8:42 pm

I got the impulses and am keen to get more non-reverb impulses for filtering.

With young kids I don't get the time to take my Zoom and Shotgun Mic out recording so this thematic collection is going to be very useful.

unmode
Common Wiggler
Posts: 69
Joined: Wed Feb 05, 2014 10:05 am

Post by unmode » Thu May 15, 2014 6:39 am

Can highly recommend those spirit canyon IR packs, if you can track them down now that the site has vanished.

My favourite one is Spectral Relativity. Some of the really big and fancy ones can turn a couple of saw waves into a whole orchestra. They will cause havok with the phase of your overall mix though. I generally use the insane ones as compositional tools and then eventually replace or remove them later on.

User avatar
tilman
Common Wiggler
Posts: 109
Joined: Thu Oct 06, 2011 8:56 am
Location: Germany / Cologne

Post by tilman » Thu May 15, 2014 11:26 am

Have those Spectral Relativity too - very usefull!

User avatar
tremblap
Common Wiggler
Posts: 216
Joined: Fri Sep 20, 2013 1:41 pm
Location: Huddersfield, England

Post by tremblap » Mon May 19, 2014 4:19 am

Hey thanks for the props on the HISStools' Impulse Response Toolkit! We have worked hard on this, and it is available for free for Max!

Just so you know, there was talk of porting to Pd and to Supercollider, as well as standalone apps for Mac and PC (in form of plugin as well) so let me know if I should post here about them.

talk soon

pa

User avatar
maurice
Common Wiggler
Posts: 167
Joined: Tue Jan 07, 2014 7:20 am
Location: Paris, FR

Post by maurice » Wed Jun 11, 2014 10:55 am

madronalabs wrote:Convolution can only reproduce the parts of audio processes that are linear time invariant (LTI).
There is a way to model anything through convolution, using Hammerstein cascades : you model any transfer fonction by a polynom, first order is linear convolution, then convolution with 1st harmonics, 2nd ...

I have no idea if any software does this, but it is used to model non linearities in big devices like wavefield synthesis or microphone arrays up to 4th order

User avatar
tremblap
Common Wiggler
Posts: 216
Joined: Fri Sep 20, 2013 1:41 pm
Location: Huddersfield, England

Post by tremblap » Wed Jun 11, 2014 1:19 pm

In the hiss tools there is an object to decline the different Hammerstein impulses...

User avatar
maurice
Common Wiggler
Posts: 167
Joined: Tue Jan 07, 2014 7:20 am
Location: Paris, FR

Post by maurice » Thu Jun 12, 2014 8:29 am

Exactly ! So using sinesweeps and HISS software you should be able to get a better approximation of non linear behaviors.


the documentation of HISS reads :party:

irnonlin
∼
the non-linear components retrieved from
an ESS measurement relate to the individual harmonics
of a system. If these are transformed appropriately, it be-
comes possible to model the non-linear system by a Ham-
merstein model which operates on consecutive powers of
the input signal, rather than harmonic transposi-
tions. N convolutions are necessary to model the first N
harmonics. The irnonlin object, takes N input IRs for
the harmonics (taken with the irmeasure object) and returns N new IRs for use in a Hammerstein model.

peripatitis

Post by peripatitis » Thu Jun 12, 2014 7:06 pm

tremblap wrote:Hey thanks for the props on the HISStools' Impulse Response Toolkit! We have worked hard on this, and it is available for free for Max!

Just so you know, there was talk of porting to Pd and to Supercollider, as well as standalone apps for Mac and PC (in form of plugin as well) so let me know if I should post here about them.

talk soon

pa
Had no idea you were related with this.
Congratulations :tu: , the toolkit is really great and it was something really missing in max !
Great to see you will port it to other platforms as well.

Any plans to allow for real-time grabbing of response ?
So far i've only seen kyma do that.

What i usually do is use a matrix of responses with the same signal input and cross-fade between them and i thought perhaps it would be cool if one could 'scan' the response buffer instead.
Don't know if it is possible though :)

Any way, really great stuff !!!

User avatar
madronalabs
Common Wiggler
Posts: 210
Joined: Sat Mar 19, 2011 1:31 pm
Location: Seattle
Contact:

Post by madronalabs » Fri Jun 13, 2014 2:34 am

There is a way to model anything through convolution, using Hammerstein cascades
Understood, I was trying to keep things simple and explain something about basic convolution.

I have a basic grasp of what Volterra / Wiener series do, but not a solid understanding of the math. I can see how responses of different harmonics could be captured by different kernels, using Taylor series as an analogy. On the other hand, even low-order nonlinearities can lead to chaotic behavior, which my intuition tells me can't be reproduced with this kind of method. Is my intuition wrong? Or if not, can anyone explain what kinds of behavior are left out?

User avatar
tremblap
Common Wiggler
Posts: 216
Joined: Fri Sep 20, 2013 1:41 pm
Location: Huddersfield, England

Post by tremblap » Fri Jun 13, 2014 9:29 am

peripatitis wrote:Had no idea you were related with this.
Congratulations :tu: , the toolkit is really great and it was something really missing in max !
My pleasure ;-)
peripatitis wrote: Great to see you will port it to other platforms as well.
Don't hold your breath ;-) The C code has been released and there was talk with the Pd community to port it but I don't know where this project is at...
peripatitis wrote: Any plans to allow for real-time grabbing of response ?
So far i've only seen kyma do that.
What do you mean? There are many ways to do this, and we have two measurements project plus buffer~ processing ones so if you tell me how it works in Kyma then I can maybe help, here or via the email in the help files.

p

User avatar
maurice
Common Wiggler
Posts: 167
Joined: Tue Jan 07, 2014 7:20 am
Location: Paris, FR

Post by maurice » Mon Jun 16, 2014 10:06 am

madronalabs wrote:
There is a way to model anything through convolution, using Hammerstein cascades
Understood, I was trying to keep things simple and explain something about basic convolution.

I have a basic grasp of what Volterra / Wiener series do, but not a solid understanding of the math. I can see how responses of different harmonics could be captured by different kernels, using Taylor series as an analogy. On the other hand, even low-order nonlinearities can lead to chaotic behavior, which my intuition tells me can't be reproduced with this kind of method. Is my intuition wrong? Or if not, can anyone explain what kinds of behavior are left out?
Exactly, it's mostly a Taylor polynomial approximation. If you apply a polynomial function to a sinesweep, due to Fourier magic the different kernels are exactly visible in the time frequency plane : the degree two is the action on the first harmonic, and so on.

Roughly, 4-5 orders polynoms can model a large family of functions, including a decent approximation of compression, or any kind of thresholding. It works well for speakers (even the cheap socalled non linear ones), microphones, room equalisation. So you get 4-5 convolution kernels, which you apply on different multiples of your signal.

What you will always lose is the noise, the spikes/crackles, vibrato, anything truely chaotic, so possibly the most interesting part when sampling badass gear.

User avatar
madronalabs
Common Wiggler
Posts: 210
Joined: Sat Mar 19, 2011 1:31 pm
Location: Seattle
Contact:

Post by madronalabs » Wed Jun 18, 2014 10:33 pm

Thanks for the info, maurice.

Post Reply

Return to “Music Software”