What's new

Convo Reverbs - What's the Diff?

SchnookyPants

I never metaphor I didn't like
As I bask here in my ignorance I'm thinkin'... "Given the same impulse-response, shouldn't all convo reverbs sound identical?"

Please enlighten me. If convo units A and B are both using the ir called, "Super Cathedral 3000 at 50% Humidity", what kind of things make A sound different than B?
 

shawnsingh

Active Member
From what I know about DSP, even though convolution is a well defined math operation, in practice the implementation uses FFT + simple multiplication + inverse FFT, because it's much faster computationally. But then there are a lot of details that could be different in how those FFTs are implemented. But still, my guess is that those implementation variations, if done correctly, shouldn't result in different sounding audio. Maybe I'm wrong about that?

Have you actually found the same IR in two different plugins produces different results? My first guess is that there's additional processing like EQ filters or cutoffs that would explain the differences. Or maybe one was 4-channel true stereo and the other was only 2-channel.
 

Manuel Stumpf

Active Member
If the convolution reverbs strictly apply the so called "convolution" as defined in mathematics and use the exact same impulse response without any other modifications, the outcome indeed are exactly the same (with slight rounding errors due to different software programming underneath). Except if they have messed up the mathematics behind it :laugh:.

Many convolution reverbs however do additional things:
Setting different pre delays.
EQing different frequency spectrum ranges.
Some allow setting the decay time of the reverb (which somehow has to modify either the impulse response before the convolution, or does some other strange processing).
Or cut the impulse response into early reflections and tail, which can be adjusted independently.
Or do some time warping on the impulse response, or let you dial in the "size" of the room etc.
All these things make the reverbs sound different.

Second thing is developers claiming to claiming to have the impulse response of the same room, can be very different:
When buying a convolution reverb a big junk is the impulse responses you buy not the algorithm itself.
Making good impulse responses is like sampling.
Even if two developers sample the same room, the outcome can be different due to different microphone setup, positioning, used algorithms to extract the impulse response from their measurements, denoising, post processing, how they deal with mono/stereo/surround...

Strictly speaking every combination of instrument location (sound source) in the room and position of the listener (sound receiver) would lead to a different impulse response.
Therefore the room sounds different from different spots of listening/instrument placement.
In a perfect world you would have one impulse response for each instrument multiplied by listening location. Which is totally impractical to handle.

It's like buying two string libraries then finding out they sound totally different,
despite they both have sampled the same instrument.

Edit: Removed some typos
 

Nick Batzdorf

Moderator
Moderator
If you search for "Cholakis," you'll find the answer to this question as well as a lot of interesting information about convolution.

Ernest Cholakis' company is Numerical Sound. He's a purveyor of custom impulse responses, among other things.
 

Manuel Stumpf

Active Member
Ah. For a moment forgot about that one :).
Probably I was distracted by diving around in the new Synchronized Special Edition :laugh:.

Is the spatialisation purely relying on convolution with impulse responses?
I have the feeling Vienna MIR Pro is doing a lot more than just convolution under the surface (just guessing by the amount of parameters you can adjust).
It is the basic principle behind it of course.
 

shawnsingh

Active Member
Ah. For a moment forgot about that one :).
Probably I was distracted by diving around in the new Synchronized Special Edition :laugh:.

Is the spatialisation purely relying on convolution with impulse responses?
I have the feeling Vienna MIR Pro is doing a lot more than just convolution under the surface (just guessing by the amount of parameters you can adjust).
It is the basic principle behind it of course.
Had some conversation with Dietz in this thread which may shed light to you on what additional processing MIR does. For one, impulse responses aren't just for positioning, they also characterize the orientation and directivity pattern of the source recorded impulse. As I understand, MIR pro rooms have sampled multiple orientations in addition to so many positions across the room. Also, it has a mathematically correct way of representing different microphone polarity patterns that you can configure in detail. The, there's the question of how to map a virtual instrument's stereo samples to the impulse responses in the room, and to do that, MIR is modelling the width, orientation, and directivity pattern of individual instruments.

Disclaimer - I like to be nerdy about audio DSP but don't really have implementation experience with these things. Forgive me if I get details wrong =)
 

Dietz

Space Explorer
Is the spatialisation purely relying on convolution with impulse responses?
The "reverb" part relies on convolution in every aspect. (The algorithmic reverb add-on "MIRacle" doesn't count, it's just for sweetening purposes).

MIR's source positioning is based Ambisconics _and_ positional impulse responses (... which were recorded in this format, too).

In addition, the so-called "Directivity Profiles" will influence the way an instrument interacts with the room from a certain position.

... a bit of in-depth information is available in this little folder called "Think MIR":
 

Attachments

Last edited:

sinkd

Senior Member
If you search for "Cholakis," you'll find the answer to this question as well as a lot of interesting information about convolution.

Ernest Cholakis' company is Numerical Sound. He's a purveyor of custom impulse responses, among other things.
+1 for Ernest Cholakis' IRs and approach. I use them in Vienna Suite, although he has recommended LiquidSonics Reverberate as well.
 

SBK

Active Member
From my little experience in audio world, I guess every algorithm that makes the IR into an actual useful and working reverb , is different from company to company. Or maybe not. I shed a lot of light!
 

robgb

I was young once
If you want a convolution reverb that is EXTREMELY customizable, get Melda Production's ConvolutionMB. I picked it up on sale a couple months back for about $29 and it's pretty amazing, allowing you to adjust parameters you didn't even know existed, plus it's multi-band, so you can put different IRs on different frequencies. It also allows you to create your own IRs.

They also have a free version that allows you to load any IR, but has far, far less control.

https://www.meldaproduction.com/MConvolutionMB

 
Last edited:

Nick Batzdorf

Moderator
Moderator
+1 for Ernest Cholakis' IRs and approach. I use them in Vienna Suite, although he has recommended LiquidSonics Reverberate as well.
Some of his IRs are very good - I have his Hollywood Impulse Responses.

But while I have no objection to bumping him up, that wasn't really what I was intending to do. It was more to answer the original question about the differences between IR processors.

MIR is another whole story. I enjoyed playing with it a lot while I had it a few years ago.
 
Top Bottom