If I get it right a convolution reverb works like this:
Take sample from input and apply the convolution algortithm to the first sample of the IR, then on the second an so on. Repeat that for every sample of the input. So the length of the output is the input length plus the IR length.
But I want to imprint the sound characteristics of a one-shot sample on the whole input. The idea is to use a long (pad-like) sound and turn it into a short (pluck- or drum-like) sound. So we have this procedure:
Take first sample from input and apply the convolution algortithm to the first sample of the IR, then take the second sample from the input for the the second an so on. Now the output is as long as the IR unless the input is shorter than the IR. In the latter case the output will end abruptly.
Is this possible with an FX plugin? Which one.
Are there any samplers or wave editors that allow you to do this?
I would put a noise gate on the pad track (like ReaGate), set up a second track with the short plucky sound you want, the sidechain the plucky track to the gate on the pad track.
Ok, yes. The sound source applied to the "carrier" isn't supposed to be an IR. I'm thinking of cymbales and the like.
Quote:
Originally Posted by drtedtan
I would put a noise gate on the pad track (like ReaGate), set up a second track with the short plucky sound you want, the sidechain the plucky track to the gate on the pad track.
I don't want to imprint the overall volume envelope, but the tonal characteristics with it's changes in time.
Do you mean you want to play a pad chord and just trigger it with other samples to come thru?
If yes, drtedtan is right. You should use a gate to close the audio (pad sound) and use the gate per audio- or midi-input to trigger the pad signal.
If you just want the characteristics of the pad to come thru a drum sample or something like that, you maybe should use a vocoder. No idea how your idea exactly looks like.
If you just want the characteristics of the pad to come thru a drum sample or something like that, you maybe should use a vocoder. No idea how your idea exactly looks like.
A vocoder is more like it. But it would have to have quite a large number of bands (like the 100 MVocoder offers).
I thought that convolution would result in a smoother output. But it seems that I get the term wrong. OTOH, it says "mostly used for reverbs" so ther must be other uses for it
"In electronic music convolution is the imposition of a spectral or rhythmic structure on a sound. Often this envelope or structure is taken from another sound. The convolution of two signals is the filtering of one through the other."
you could do what drtedtan suggests, PLUS load an instance of ReaVerb and load your impulse file into that. While ReaVerb is provided to allow the loading of reverb IRs, and thus provide convolution reverb, it'll accept other wavs. Some of our more avant-garde forum members have loaded cymbal sounds etc and been happy with the results, or at least with the experimentation.
The gate would give you the duration portion of your request, while the ReaVerb instance would give you the tonal overlay portion of your request.
"In electronic music convolution is the imposition of a spectral or rhythmic structure on a sound. Often this envelope or structure is taken from another sound. The convolution of two signals is the filtering of one through the other."
Create the sound with the spectral and rhythmic structure you want, load it in a convolution plugin like you would a reverb IR, then run the sounds you want to impose that spectral and rhythmic structure on, through the convolution plugin.
Take first sample from input and apply the convolution algortithm to the first sample of the IR, then take the second sample from the input for the the second an so on. Now the output is as long as the IR unless the input is shorter than the IR. In the latter case the output will end abruptly.
Is this possible with an FX plugin? Which one.
Are there any samplers or wave editors that allow you to do this?
Masi
Top
Sounds like you're describing Amplitude Modulation or Ring Modulation. It will give you Sum and difference harmonics of both sounds. Kind of noisy and distorted sounding with a non harmonic sound. Don't know of any plugs that do this but there probably is one somewhere.
Create the sound with the spectral and rhythmic structure you want, load it in a convolution plugin like you would a reverb IR, then run the sounds you want to impose that spectral and rhythmic structure on, through the convolution plugin.
Done.
Unfortunaltely not. I have not convolution plugin that doesn't work as a reverb. I don't want the samples of the input source to be convolved by the whole spectral/rhythmic source. I would need to setup an IR reverb for every sample of the spectral/rhythmic source.
So the suggestion to use a vocoder instead is good. I gave it a try, but the result doesn't convince me. But maybe the whole idea is just dumb.
Well sure, if you use some sort of percussion sample as the IR and run a pad through it you'll get some funky unrealistic kind of reverb.
If you want to impose the dynamics of the percussive track onto your pad sound, then some kind of gating using the percussive on the sidechain (or parameter mod of volume?) is a way to go. If you want to impose both the dynamics AND the spectral quality of the percussive sound on the pad, then a vocoder might well work - ReaVocode does up to 144 bands.
You might get the most satisfying effect with a mix of gating and vocoding - I don't know how ReaVocode time constants are set - i.e. you can't control the attack and release, so you might well be losing the attack a bit - you could make this up with a short attack / release gated version alongside. Of course, if your vocoder has the bands and access to the attack time you might not need to do anything else.
__________________ it's meant to sound like that...
But I want to imprint the sound characteristics of a one-shot sample on the whole input.
So the problem seems to be in defining what is the 'sound characteristics' of the one-shot sample.
Quote:
Now the output is as long as the IR unless the input is shorter than the IR. In the latter case the output will end abruptly.
Your definition here of 'sound characteristics' seems to be the length of the one-shot sample.
I'm not pokin' fun or anything like that. I'm pointing out that the first problem is specifically defining the problem to be solved; i.e., what specific aspects of the one-shot do you want to imprint upon the input?
__________________
It's time to take a stand against the synthesizer.
Last edited by brainwreck; 07-21-2018 at 07:01 AM.
Is there something that could be done with ReaFIR that would come close? Capture the spectrum of the "sample" and then run your pad through it. I guess that won't follow the dynamics of the percussive thing, but we have other ways of accomplishing that.
There was a Pluggo plugin called Convolver that could kind of do it with a pair of live inputs, but I wouldn't suggest messing with those plugs nowadays.
BTW - I have a naive but useful JS Ring Modulator in the stash. Send your mono carrier on one channel and the modulator on the other. Doesn't matter which is which. Mono output. If you want stereo, you can use two of them.
I'm not pokin' fun or anything like that. I'm pointing out that the first problem is specifically defining the problem to be solved; i.e., what specific aspects of the one-shot do you want to imprint upon the input?
Seems so.
So many kind people here trying to help me
... and I'm failing to describe the problem properly.
Thanks to all of you!!!
I think that those who suggest a vocoder understand what I want. Perhaps another way to describe the idea is a real-time matching-EQ. I want to create a crash that has its frequency content replaced by the pad sound.
I'll try to play the crash through ReaVerb but loading a very short sample of the pad as the IR.
No worries. If we didn't have an interest in this stuff, we wouldn't be participating in the discussion.
Quote:
Originally Posted by Masi
I want to create a crash that has its frequency content replaced by the pad sound.\
Can you describe what the end result might sound like? Should the overall timbre of the crash remain intact, or should it be altered in some way? Should the volume of the crash over time remain intact...?
__________________
It's time to take a stand against the synthesizer.
the IR determines what it actually is...
(ie...you can load an EQ IR and then it becomes an EQ. etc.)
That's not true. The whole IR (all samples) will be played for each single sample of the source *. If the IR is short enough you won't hear it as a reverb though.
Masi
* I still think that the way I described a convolution reverb in my OP is correct.
i know how it works, and the length of the IR has nothing to do with anything...i can make a 10 second long EQ/filter IR, so does that mean its a reverb when i load it into ReaVerb ? no, it's an EQ/filter.
ReaVerb does convolution. it can work just like Q-clone. an IR is an IR.
I have not convolution plugin that doesn't work as a reverb.
I was not correct. After all I do own plugins like that. One is Boogex, the free guitar amp by Voxengo. It's cabinet simulator * is a convolver that let's you load custom IRs.
The other one is Trash 2 by iZotope.
I use them now to load a bit of my pad sound as IR and play the crash through it. Sounds much better then abusing a convolution reverb. With ReaVerb I found it hard to find the right length for the IR. But that may depend on the sound.
Using all 144 bands of ReaVocode was ok. Though even with this many bands it didn't deliver the sound I was after. It lost too much of the crashes characteristics. MVocoder with 100 bands and att/rel time of 0.1ms was better.
Masi
* The JS "Convolution Amp" does the same but you have to tweak it to use custom samples (or drop your samples in the correct folder). Another free convolution cab sim is NadIR by Ignite Amps.
I use them now to load a bit of my pad sound as IR and play the crash through it. Sounds much better then abusing a convolution reverb. With ReaVerb I found it hard to find the right length for the IR. But that may depend on the sound.
Just curious here. Can you post a before and after convolution example?
__________________
It's time to take a stand against the synthesizer.
i know how it works, and the length of the IR has nothing to do with anything...i can make a 10 second long EQ/filter IR, so does that mean its a reverb when i load it into ReaVerb ? no, it's an EQ/filter.
ReaVerb does convolution. it can work just like Q-clone. an IR is an IR.
I don't know the details on the workings of convolution. What is the difference in how a short and long impulse response are treated? In other words, what makes a 10 second room impulse response carry out over time, sounding like a reverb vs. a 10 second cab impulse response, which according to what you are saying, would not sound like a reverb? I only know that cab impulse responses tend to be very short in comparison to room impulse responses. And so from my perspective of ignorance on the workings of convolution, it seems that impulse response length is the deciding factor in what an impulse response does.
__________________
It's time to take a stand against the synthesizer.
First sound is the section of the flapping sound I used. Next is the test signal, then the test signal with the flap sound as an IR in ReaVerb. Finally I send the flap sound through itself just for the hell of it.
I don't know the details on the workings of convolution. What is the difference in how a short and long impulse response are treated? In other words, what makes a 10 second room impulse response carry out over time, sounding like a reverb vs. a 10 second cab impulse response, which according to what you are saying, would not sound like a reverb? I only know that cab impulse responses tend to be very short in comparison to room impulse responses. And so from my perspective of ignorance on the workings of convolution, it seems that impulse response length is the deciding factor in what an impulse response does.
My understanding is along the following lines.
An IR of a room is more accurate the longer it takes to play the sweep, so longer is preferable, but that also opens the door for issues like passing cars etc ruining the take.
That's separate from the actual response - a 10-second reverb IR would be a pretty big room - it would take a full 10 seconds for all the reflections to disperse. But it could be captured via either a 3 second sweep or a 30 second sweep.
I would think that guitar cab responses are more stable over time, so a shorter sweep is just as good as a longer one.
But I don't know for sure, that's just putting together what I do know in a way that fits.
In other words, what makes a 10 second room impulse response carry out over time, sounding like a reverb vs. a 10 second cab impulse response, which according to what you are saying, would not sound like a reverb?
why does 10 seconds of piano sound different than 10 seconds of guitar ?
First sound is the section of the flapping sound I used. Next is the test signal, then the test signal with the flap sound as an IR in ReaVerb. Finally I send the flap sound through itself just for the hell of it.
An event in one input get stretched to the complete non-silence duration in the other.
Well, it's actually exactly like triggering the entire impulse "sample" over and over again for every sample that comes in the input. A single sample step function input will sound exactly like the impulse file played all the way through.
I'm sure that ReaFIR or something like it is as close as the OP will get. I'm also sure that the artifacts inherent in that type of resynthesis process will be tough to deal with.
i know how it works, and the length of the IR has nothing to do with anything...i can make a 10 second long EQ/filter IR, so does that mean its a reverb when i load it into ReaVerb ? no, it's an EQ/filter.
ReaVerb does convolution. it can work just like Q-clone. an IR is an IR.
You are giving too much importance on the source of the IR. It doesn't matter if the IR is recorded in a real space. It could be completely synthisized. What it makes sound like a reverb is that over some time the entire spectrum is getting more silent.
And I think the term IR stem from the fact that a single impulse (a short sound) is generated to record a possibly long response (think cathedral).
So anything that fades away, like a 5sec cymbal recording, used within a convolution reverb will sound as a reverb. A static 5sec synth pad will used the same way will not be recognized by our brain as a reverb.
And I think the term IR stem from the fact that a single impulse (a short sound) is generated to record a possibly long response (think cathedral).
No, "impulse response" is the mathematical concept that describes how a linear time-invariant system (like a reverb) affects a signal passing through it.
In fact, usually an impulse is NOT used to sample impulse responses, but a sweep, a chirp, or a similar wide-spectrum signal.
Convolution is a mathematical operation between two signals, and it allows to reproduce the effect of a finite-impulse-response (FIR) system on a signal, without actually implementing the system but instead using its IR as a source of "mathematical description" of the system.
If you want convolution to work differently, you DON'T want convolution.
If I understood you correctly, you want modulation.
If I understood you correctly, you want modulation.
Quite possibly. More and more people are telling me that I got it wrong and I tend to believe them
Unfortunately termks like "linear time-invariant system" and "finite-impulse-response" are only empty words for me. My math is way too poor to understand them.
Linear means no compression/expansion/distortion - the output does change based on the level of the input.
Time-invariant means no modulation - a given input will create the exact same output no matter what time it is.
FIR also means pretty much exactly what if means. The impulse response has a finite length. It has a definite beginning and end. Like .wav files that we use in our convolution filters, but there are other ways to do it also. This as opposed to an IIR like a "normal" filter that basically just keeps dividing the original impulse sample down to create its output and theoretically won't ever reach zero. Most real world things are actually IIR, but we can get way with representing them with FIR because eventually an infinite response decays into the noise floor and becomes meaningless anyway.
Time-invariant simply means that the system doesn't change its behaviour in time. Whenever you feed the exact same input to it, you always get the exact same output (so, for example, NOT a flanger or "tremolo" or any other effect with an LFO).
Linearity is a more math-heavy concept. Let's say that if you feed a system with input X, you get output W, and if you feed it with input Y, you get output Z. The system is linear if the output for a*X+b*Y is exactly a*W+b*Z. This implies as a particular case that the output for a*X is a*W, or, put in more practical terms, if you boost the input signal by 6dB, the output signal is boosted by 6dB but doesn't change in any other way. This makes it easy to see that distorsion and compression are typical examples of non-linear systems.
Side note: in reality, it's practically impossible to find PERFECTLY linear and time-invariant physical systems, but since these two properties make MASSIVELY easier to make a mathematical model or analysis, systems get simplified to linear time-invariant whenever possible (that is, whenever the non-linearities and time variations are either not desirable, or small enough to be basically irrelevant).