what about resynthesis techniques applied to impulse responses ?

[Krakatau]

This is just an idea inspired by the use of raw samples loaded as IR into convolution engines

Usually, it takes time to have a usuable result, often just using a specific sample as IR with a specific sound source

By experience,i saw that in the frequency domain (?) a main issue is related to the fact that a specific one (more often a bandwith, though...) is drastically enhanced when beeing present in both soundsource and IR file and this would change completely just by playing a different tone in your soundsource instrument

In order to recover a relative predictability, you just need to tune the IR file (as any sample during sample editing), that what a rompler as Galaxy X seems to offer as a new feature :

www.galaxypianos.com/galaxy-x.html

Of course, it would make sense composition-wise only to match (often rather roughly) a choosen tonality

But in such case, if the IR datas could be for instance rendered like in the image synth room of metasynth 5 in where a sample could be directly rendered as beeing filtered by a choosen tonality (as a filtering canvas)...it would be then quite more useful and easier to handle as a kind of sympathetic raisonnance, focused (but limited) on the choosen tonality

Then eventually mixed or in the case resynthesis dates morphed with the original sample to have an equilibration between an original sample and its tuned/harmonised version as impulse response

But this is just an example, among other possibilities of making IRs more flexible to use in general ...more that the envelope and time strech functions that currently exists for such tasks


I assume that, as i do, many sound designers have experienced this by combining different applications

I suppose though, that the current platforms are or will be very soon able to handle such tasks for a real time use so i thought perhaps it might make sense to start this thread for suggestions and critiques

[Krakatau]

Sorry, probably this topic belongs to the sound design subforum rather than this one ..?

[quintosardo]

Even though not through IRs but through delay networks, I think it is used in some piano emulations? (tune the delays instead of building a chord IR, very similar)
I think that the more you restrict the resonances, the more the reverb sounds metallic.
Experimenting with it a lot, here.

[valhallasound]

Google "commuted waveguide synthesis." The idea is that, in any instrument that can be expressed as a waveguide model being processed by a complex linear resonator, you can use the impulse response of the resonator as the excitation for the waveguide and get the same sound.

This works for struck/plucked stringed instruments: guitar, piano, etc.

Of course, sampling works even better for piano, once the streaming sample technology was developed. And lots of people play guitar. And convolution with a complex impulse is easier nowadays. So commuted waveguide synthesis was a research idea that peaked in the late 1990's, but never seemed to go very far.

Sean Costello

[Krakatau]

Though your comments are highly instructive, the parrallel i would make with tuned delay lines is a bit different

i'm a big fan of these unproperly called "comb filter" when AFAIK the delay time of a feedback loop is identical to a choosen frequency

For instance a delay line at 25 ms and a fundamental frequency at 40 hz
(if i'm not wrong : nobody at the present time did confirm or not what i suspected by instinct)

By comparison, here a few IRs created from scratch that intend to recreate similar drones effects on a G major scale and all it modal transpositions

(with the qualities and defects inherent to the convolution technique)

www.anak-krakatoa.net/~downl/misc/G.exe

In practice you can perfectly place them on a send/return loop of your mixing table at the time your arrangement matches the choosen scale, (that was the basic idea in the case)

What i suggested in my first post would be a way to quickly sculpt your impulse response in a way it would similarily match a choosen scale/tonality without alienating too much the feel induce by the original IR (after some tuning and/or other adjustments if needed)

I apologise because i wont be back on this topic before monday and i regret it, just give a try to these IR as suggested if you feel so...

[espiegel123]

I personally haven't had a chance to explore this (time being finite), but a couple of sound designers have contacted me in the past and told me that they use the MetaSynth Image Synth and Image Filter rooms pretty extensively in processing and/or synthesizing impulse responses for use with Altiverb to create wild IRs. I would think that the Spectrum Synth could create some pretty interesting IRs.

This is just an idea inspired by the use of raw samples loaded as IR into convolution engines

Usually, it takes time to have a usuable result, often just using a specific sample as IR with a specific sound source

By experience,i saw that in the frequency domain (?) a main issue is related to the fact that a specific one (more often a bandwith, though...) is drastically enhanced when beeing present in both soundsource and IR file and this would change completely just by playing a different tone in your soundsource instrument

In order to recover a relative predictability, you just need to tune the IR file (as any sample during sample editing), that what a rompler as Galaxy X seems to offer as a new feature :

www.galaxypianos.com/galaxy-x.html

Of course, it would make sense composition-wise only to match (often rather roughly) a choosen tonality

But in such case, if the IR datas could be for instance rendered like in the image synth room of metasynth 5 in where a sample could be directly rendered as beeing filtered by a choosen tonality (as a filtering canvas)...it would be then quite more useful and easier to handle as a kind of sympathetic raisonnance, focused (but limited) on the choosen tonality

Then eventually mixed or in the case resynthesis dates morphed with the original sample to have an equilibration between an original sample and its tuned/harmonised version as impulse response

But this is just an example, among other possibilities of making IRs more flexible to use in general ...more that the envelope and time strech functions that currently exists for such tasks


I assume that, as i do, many sound designers have experienced this by combining different applications

I suppose though, that the current platforms are or will be very soon able to handle such tasks for a real time use so i thought perhaps it might make sense to start this thread for suggestions and critiques

[zynaptiq]

MetaSynth is indeed excellent for creating wild IRs (amongst a whole bunch of other things).

There's also a module in Kyma called "R/E-Synthesis" (short for "Resonator-Excitator"), which does the commuted waveguide thing....it separates any sample you throw at it into an excitator file (basically amplitude-variant pulses) and a complex, time-varying FIR filter topology (a sequence of impulse responses, basically).

Works pretty well for doing MAD stuff that still sounds somewhat natural.

Then there's the Hartmann Neuron, which does a vaguely similar thing, though based on a different technology. It reverse engineers a physical model from a sample, then lets you modify the excitator separately from the corpus, and combine these two entities (called "Scape" and "Sphere") from two different models. It isn't convolution based, but from a workflow methodology point of view it's a similar thing.

[espiegel123]

Interesting that you mention the Hartmann Neuron. One of the guys developing presets for them a number of years ago was using MetaSynth for many of the sounds and remarked that the two of them had great synergies when used together.

MetaSynth is indeed excellent for creating wild IRs (amongst a whole bunch of other things).

There's also a module in Kyma called "R/E-Synthesis" (short for "Resonator-Excitator"), which does the commuted waveguide thing....it separates any sample you throw at it into an excitator file (basically amplitude-variant pulses) and a complex, time-varying FIR filter topology (a sequence of impulse responses, basically).

Works pretty well for doing MAD stuff that still sounds somewhat natural.

Then there's the Hartmann Neuron, which does a vaguely similar thing, though based on a different technology. It reverse engineers a physical model from a sample, then lets you modify the excitator separately from the corpus, and combine these two entities (called "Scape" and "Sphere") from two different models. It isn't convolution based, but from a workflow methodology point of view it's a similar thing.

[zynaptiq]

I believe that might have been me

[espiegel123]

I believe that might have been me

Hey D, Fancy meeting you here. Believe it or not, it was someone else that I had in mind. I just looked up the message he sent a few years ago to be sure. I don't think I have permission to "out" him, though.

-Edward

[synthgeek]

Bidule is my favorite program for experiments along these lines, because it includes a realtime convolution component, and lots of fun spectral toys.

Not being a Mac user, I know nothing about MetaSynth... however, looking at the info on the image synth on their website, I think Bidule might be capable of results along the lines you're thinking of in the OP (if I'm understanding properly) - though obviously the "experience" would be different due to different interfaces and working methods.

[synthgeek]

Here's a Bidule group inspired by the "tonal filter" idea in the OP (in case anyone else here uses it):

http://synthgeek.skincontact.com/files/ ... trocon.zip

Note: I'm not sure how true this is to your vision, Krakatau, in fact I don't think it's much like what you're talking about doing in MetaSynth, but it's sort of a mix of what I got out of your description, and the basic idea of tuned IR's.

It has two stereo inputs for the audio to be processed - the first pair just goes right to the convolution modules, but the second goes through 3 FFT-based spectral gates first. The "pitch" (center frequency) of the gates is MIDI controlled, and the 2nd & 3rd gates can be transposed from the first by +/- 24 semitones (the main pitch can be transposed too).

There's a lot more that could be done with this using Bidule's spectral tools, but this was done as a quick mod of something else I recently cleaned up and somewhat organized for public consumption:

http://synthgeek.skincontact.com/files/sg-filtcon.zip

It's a much simpler version of a similar idea, basically just a filter + convolution. It was originally just something that came about playing with (semi-)tuned percussion.

[Krakatau]

Shortly and roughly

My idea is to add to the convolution reverb an effect stage applied to the impulse response sample (or that might create an IR sample from scratch) before loading it into the convolution engine

I apologise, i'll need to find the time and take my head on the few previous post to see exactly what's going on, i'm afraid i'm a bit busy yet

[synthgeek]

My idea is to add to the convolution reverb an effect stage applied to the impulse response sample (or that might create an IR sample from scratch) before loading it into the convolution engine

Ok then, that much I got right. Bidule's convolution engine is just a simple 2-input/1-output real-time module, so there's nothing to load per se... though you could certainly load a wave file into one of the players to use it as one of the signals.

The main thing I'm not clear on as far as your MetaSynth description is whether it would be filtering the sound or pitch shifting it... or maybe something else?

I apologise, i'll need to find the time and take my head on the few previous post to see exactly what's going on, i'm afraid i'm a bit busy yet

No worries.