convolution based emulations
-
kensuguro
- Posts: 4434
- Joined: Sun Jul 08, 2001 4:00 pm
- Location: BPM 60 to somewhere around 150
- Contact:
convolution based emulations
I'm surprised I'm not finding any plugins aimed specifically at emulating gear using convolution.. exactly what focusrite is doing with their liquid series. Seems like a logical way to do emulating without meticulous modeling.
a couple of years ago I really thought convolution is a very interesting approach to 'model' certain types of gear.
now after dealing with more analog signals, guitar and and bass and owning a tube amp in physical form, various speakers and (occasionally strange) outboard gear, different preamp stages, passive parts etc. I'm not that sure anymore.
but thanks for your question - it triggered an interesting idea...
cheers, Tom
now after dealing with more analog signals, guitar and and bass and owning a tube amp in physical form, various speakers and (occasionally strange) outboard gear, different preamp stages, passive parts etc. I'm not that sure anymore.
but thanks for your question - it triggered an interesting idea...
cheers, Tom
My Brothas',
I realised years back when I first got Scope that trying to compare, or replace analog hardware could only get close at best. But what I did realise is how to use both forms, i.e. analog / virtual, to compliment each other. Try mixing a monophonic analog hardware synth with a Bowen synth, or enhancing a rotary cabinet by extending it's Doppler effect using a virtual ( Celmo's or Creamware's ) with the real thing.
The topic of which one is better is no longer valid for me, as I can't use hardware or virtual by themselves anymore, as they now seem to extend and compliment each other.
True, most emulations can pass at times in recordings, but my live quest for the best sound has also been used to great effect in recordings as well.
Gigastudio 4 will be using it's awesome GigaPulse app., which stared as a convolution reverb, and resonant body emulator of acoustic instruments. Emulation of different effects and instruments should be very interesting. Maybe that's why they are feverish to go to 64 bit, w/ it's added memory addressing, and multi-threaded, multi-core approach.
I Cannot Use One W/O The Other,
I realised years back when I first got Scope that trying to compare, or replace analog hardware could only get close at best. But what I did realise is how to use both forms, i.e. analog / virtual, to compliment each other. Try mixing a monophonic analog hardware synth with a Bowen synth, or enhancing a rotary cabinet by extending it's Doppler effect using a virtual ( Celmo's or Creamware's ) with the real thing.
The topic of which one is better is no longer valid for me, as I can't use hardware or virtual by themselves anymore, as they now seem to extend and compliment each other.
True, most emulations can pass at times in recordings, but my live quest for the best sound has also been used to great effect in recordings as well.
Gigastudio 4 will be using it's awesome GigaPulse app., which stared as a convolution reverb, and resonant body emulator of acoustic instruments. Emulation of different effects and instruments should be very interesting. Maybe that's why they are feverish to go to 64 bit, w/ it's added memory addressing, and multi-threaded, multi-core approach.
I Cannot Use One W/O The Other,
-
BingoTheClowno
- Posts: 1722
- Joined: Wed Nov 12, 2003 4:00 pm
- Location: Chicago
- Contact:
-
kensuguro
- Posts: 4434
- Joined: Sun Jul 08, 2001 4:00 pm
- Location: BPM 60 to somewhere around 150
- Contact:
I don't remember the details, but it's convolving one waveform with another. Usually it's a base signal and some impulse. It's sort of like using the impulse as a filter for the base signal. The impulse is a recording of a wideband sound (such as a start gun bang). The gun is fired in a room / fed through gear, etc, and then recorded. Convolution takes that recording, and reverse applies it onto the base signal. The end result is that the base signal takes on characteristics of the impulse. Most of this stuff is used for convolution reverbs, which are quite nice sounding.
But people have also recorded impulses from gear like reverb units, comps, eq, etc. and used convolution to emulate these gear. Focusrite Liquid is a hardware unit that does exactly that.
Also, like braincell pointed out, you can also get some really crazy sounds with convolution. Like with scope4live's example, play stuff through the resonant body of another instrument, etc. It only gets crazier from there.
But people have also recorded impulses from gear like reverb units, comps, eq, etc. and used convolution to emulate these gear. Focusrite Liquid is a hardware unit that does exactly that.
Also, like braincell pointed out, you can also get some really crazy sounds with convolution. Like with scope4live's example, play stuff through the resonant body of another instrument, etc. It only gets crazier from there.
I use a convolution reverb to emulate the body resonance of a stredivarious, or an acoustic piano w/ pedal down, or early reflections of acoustic spaces, then add my hardware reverb, i.e. the Q20 w/ ADAT I / O's, or the PCM91 via AES / EBU all into Scope, ala in the box. It is awesome sounding in all of it's splendor.
SIR is an excellent way to get your feet wet BTW, good link Braincell.
SIR is an excellent way to get your feet wet BTW, good link Braincell.
There is a VST plugin that does this stuff. It uses a type of dynamic convolution (not the same as Sintefex/Focusrite Liquid) so processors that act in the time domain can be 'sampled'.
http://www.acusticaudio.com/
http://www.acusticaudio.com/
Here's my brief explanation on convolution
The Wikipedia article links to a visual demonstration of convolution.
You can actually see there how applying a lowpass filter over a square wave makes a mellower signal:
Each filter (or reverb or other linear time-invariant systems) is defined by an impulse response (IR) - how would its output respond to an impulse signal in its input.
A typical lowpass filter has an impulse response of a falling exponential: Meaning that if you drive an impulse in its input, the output of the filter would be a falling exponential - "a pulse that goes down slower in time".
When a signal enters in the filter, what comes out is a convolution of the signal with the filter's unique impulse response:
Output = Input * IR
or:
y(t) = x(t) * h(t)
Where * is the convolution operator.
So if you open the above demo,
for x(t) you select the square wave,
for h(t) you select the falling exponential,
then drag one over the another to convolve between the two - you see that the RESULT y(t) is a smoother square wave, with much less high-frequency contents
Now, where's the cutoff?
Suppose that an impulse response is simply an impulse (delta function). This means that the filter responses to an impulse with an impulse - hence it doesn't filter at all:
x(t) = x(t) * delta(t)
x(t) stays the same.
Not filtering at all means that the cutoff is open all the way through.
The faster the exponent of h(t) falls down, the closer it looks graphically as being an impulse, and the higher the cutoff is towards being open.
You can use the mouse in the demo to draw an shorter exponent than the provided preset - You'll see that the output signal looks more towards being similar to the input - with less filtering. (btw. the output amplitude is low because the impulse is expected to be very high)
Changing the cutoff/resonance - rapidly changes the filter's impulse response accordingly.
So how do we convert from Hz of a cutoff to an exponential slope?
This can be done using the Fourier Transform: Since the demo shows the signal in TIME domain, the properties are less intuitive. If we draw the response in the FREQUENCY domain, we can draw the transmission as it's seen in filter graphs - passing which frequency content we desire. Then we can then use the inverse Fourier Transform to go back to time domain, and then... apply convolution
A filter's transmission function (amplitude and phase as functions of frequency) is the Fourier Transform of its impulse response.
You can actually see there how applying a lowpass filter over a square wave makes a mellower signal:
Each filter (or reverb or other linear time-invariant systems) is defined by an impulse response (IR) - how would its output respond to an impulse signal in its input.
A typical lowpass filter has an impulse response of a falling exponential: Meaning that if you drive an impulse in its input, the output of the filter would be a falling exponential - "a pulse that goes down slower in time".
When a signal enters in the filter, what comes out is a convolution of the signal with the filter's unique impulse response:
Output = Input * IR
or:
y(t) = x(t) * h(t)
Where * is the convolution operator.
So if you open the above demo,
for x(t) you select the square wave,
for h(t) you select the falling exponential,
then drag one over the another to convolve between the two - you see that the RESULT y(t) is a smoother square wave, with much less high-frequency contents
Now, where's the cutoff?
Suppose that an impulse response is simply an impulse (delta function). This means that the filter responses to an impulse with an impulse - hence it doesn't filter at all:
x(t) = x(t) * delta(t)
x(t) stays the same.
Not filtering at all means that the cutoff is open all the way through.
The faster the exponent of h(t) falls down, the closer it looks graphically as being an impulse, and the higher the cutoff is towards being open.
You can use the mouse in the demo to draw an shorter exponent than the provided preset - You'll see that the output signal looks more towards being similar to the input - with less filtering. (btw. the output amplitude is low because the impulse is expected to be very high)
Changing the cutoff/resonance - rapidly changes the filter's impulse response accordingly.
So how do we convert from Hz of a cutoff to an exponential slope?
This can be done using the Fourier Transform: Since the demo shows the signal in TIME domain, the properties are less intuitive. If we draw the response in the FREQUENCY domain, we can draw the transmission as it's seen in filter graphs - passing which frequency content we desire. Then we can then use the inverse Fourier Transform to go back to time domain, and then... apply convolution
A filter's transmission function (amplitude and phase as functions of frequency) is the Fourier Transform of its impulse response.