Physical Modeling in Kontakt! (UPDATED audio example)

[Gregjazz]

and the CPU use was...? Just wondering, since you said that in Kontakt it was very low.

In SynthEdit it was pretty high. I don't remember the exact number, but SynthEdit projects tend to take a lot of CPU anyway.

[eDrummist]

We have known each other for a while and you never shared this with me! This is pretty fascinating -- a peak behind the curtain into why your libraries sound so realistic. I'm sure other sample developers are now taking notes.

I don't know a lot about Kontakt having scripting capabilities, but it appears those capabilities are far beyond what I would expect. I'm just happy to leave it to developers like you who have the skills to push the limits. All I can say is that Kontakt has meant a huge difference in terms of the realism I can achieve in my DAW based music. I look at the progress since the mid-eighties, when I started using a computer to sequence, and now, and the progress has been immense, with one exception, midi standards. Time has stood still for midi.

[Gregjazz]

Here's the latest: http://www.orangetreesamples.gregjazz.c ... aktPM2.mp3

It's a test of moving the pick position and the pickup position around. The pickup can be moved in real-time. At the moment there's only one pickup, but hypothetically you can create as many pickups as you'd like. Why settle for three single-coil pickups when you could have six? That's how rock stars think, at least.

There's still a lot to be done, but it's come a long way since the first audio example!

[freddemillio]

It continues to sound very promising. There are some interesting variations in the tone. Each note blossoms nicely.

[DaveClark]

Hi Greg,

Thanks for posting on this subject. I myself have a long-standing interest and experience in physical modelling and computational physics.

Without divulging your methods or script code, what is the model exactly? For example: What physical variable is the y-axis of the MP3's? What physical variable is being changed between the tones in the groups of five, and what physical variable is being changed between the groups?

Regards,
Dave Clark

[Download SOphist]

very interesting, will be looking forward for developments...

[dickiefunk]

Wow. I'll be watching this very closely!! Sounds extremely promising!!!

[Gregjazz]

Without divulging your methods or script code, what is the model exactly? For example: What physical variable is the y-axis of the MP3's? What physical variable is being changed between the tones in the groups of five, and what physical variable is being changed between the groups?

The model is of a string with two immovable ends, one for the bridge and one for the fret. Each end has different muting characteristics, which is the key to the harmonics dynamically changing during the note's decay. There's a pluck point, and also a pickup point (as many pickups as you want really).

There's really not too much math, it's just a practical, simple model of the way waves move on a string. It's basically the product of me holding my guitar, recording a lot of samples, studying how waves move on strings, measuring distances on my guitar with a measuring tape, and listening (this last one is VERY important).

[Jake Jackson]

Any plans to extend this by developing a soundboard model, so an acoustic instrument could be created. A very different thing, I understand, with more complex things going on across a plane. Just wondering where your thoughts are...

I'm intrigued, too, about what you said about The Trumpet using phase aligned samples. I'm not sure what is meant by aligning the phase of the samples--each sample starts either exactly in phase, or exactly out of phase to the same degree? Any links for more information? Thanks.

[Gregjazz]

Any plans to extend this by developing a soundboard model, so an acoustic instrument could be created. A very different thing, I understand, with more complex things going on across a plane. Just wondering where your thoughts are...

Sure, that's entirely possible!

I'm intrigued, too, about what you said about The Trumpet using phase aligned samples. I'm not sure what is meant by aligning the phase of the samples--each sample starts either exactly in phase, or exactly out of phase to the same degree? Any links for more information? Thanks.

They have to start exactly (or as close as possible) in phase. That way you're not essentially creating beat frequencies or canceling out frequencies when crossfading.

[DaveClark]

Hi Greg,

Thanks for your reply:

The model is of a string with two immovable ends, one for the bridge and one for the fret. Each end has different muting characteristics, which is the key to the harmonics dynamically changing during the note's decay. There's a pluck point, and also a pickup point (as many pickups as you want really).

I had already surmised as much, but thanks for confirming it.

I'm still trying to determine from what you've said whether or not what you call "physical modelling" is what physicists think of as physical modelling. It appears to me that this is not the case.

I don't know much about Kontakt at all, but:

1) The scripting language appears to be very limited for any kind of computational physics whatsoever, not being designed for this kind of thing at all.

2) It appears to me that one needs something in an instrument slot to create any kind of sound at all --- some sort of WAV for an initial condition. You've denied using any sort of WAV as I understand what you've posted.

On the other hand, if you are implicitly (e.g. with effects) or explicitly solving the wave equation with associated boundary and initial conditions then you should be able to relate your y-axis in the MP3's to some physical parameter, for example the displacement from the equilibrium position of the string at the position of the pickup when the string is plucked by some model you could describe (e.g. displacement via saw wave over 0.001 secs) at the pluck position. You should also be able to describe the vibrational modes you are considering, and so on, without giving anything away about your solution methods. In short, you should be able to discuss the physics of the problem you are modelling, not merely produce MP3's that sound like vibrating strings.

What you are doing has produced interesting results sonically; I'm interested in their relation to actual physical solutions, that is whether or not there really is a one-to-one correspondence with actual physical parameters.

Again, I'm just trying to understand how what it is you are claiming is related to what I am doing.

Still puzzled,
Dave Clark

[Gregjazz]

Hi Dave,

Let me see if I can better answer your questions. If you'd like, we could move this conversation to emails, too.

I'm still trying to determine from what you've said whether or not what you call "physical modelling" is what physicists think of as physical modelling. It appears to me that this is not the case.

Certainly there needs to be some sort of optimization of the physical model to make it more practical CPU-wise.

For example, the Karplus-Strong system of emulating the string motion is a very basic physical model of the string. Personally I think it is a great oversimplification of the process, but the Karplus-Strong method is used in many of the current physically modeled bass and guitar VSTis, since it uses so little CPU.

1) The scripting language appears to be very limited for any kind of computational physics whatsoever, not being designed for this kind of thing at all.

There is a great math plugin made by R. D. Villwock (aka "Big Bob") that really boosts the power of Kontakt scripting.

But you're correct, Kontakt is not intended for physical modeling, which is what makes my experiment exciting to explore the possibilities.

2) It appears to me that one needs something in an instrument slot to create any kind of sound at all --- some sort of WAV for an initial condition. You've denied using any sort of WAV as I understand what you've posted.

The only WAV it uses (sorry for misleading you) is a short impulse. When I used SynthEdit, this was generated on the fly, but due to Kontakt's limitations, I had to use a WAV file impulse.

What you are doing has produced interesting results sonically; I'm interested in their relation to actual physical solutions, that is whether or not there really is a one-to-one correspondence with actual physical parameters.

Back when I was using SynthEdit, I had a lot more freedom and power to process numbers. I basically would spend every day working on it, while holding my guitar and a measuring tape. I watched a lot of videos from high-speed cameras filming strings vibrating, and read up on the subject to see if I could glean useful information.

Now I got to a certain point where I do not have the tools necessary to exactly measure certain elements of the guitar, instead having to compare the generated sound with my recordings of the real guitar (I set the pickups in exactly the same position as the real guitar, and picked it in measured positions). The end result in SynthEdit, observed in a spectrum analysis, produced a sound that was nearly identical to the recorded guitar tone harmonically throughout the decay of the note. The only difference was the noise floor from the real guitar.

I don't approach physical modeling from a purely mathematical standpoint, but use my ears a lot. In my opinion, it's all about achieving a realistic generative sound in the end product, and although my physical models are based on the way strings vibrate, I do not claim that they are perfectly mathematically realistic.

[Jake Jackson]

I'm puzzled, too. But I don't Kontakt installed right now, so I can't test what happens if create code without a sample.

In any case, if no sample is needed, what happens if this method is used to add upper partials to an already complex multisample--for example, the obvious piano. Load a set of soft strike samples and across the velocity zones and layers. And then add partials this way as velocity increases? (Yes, the addition of upper partials isn't really that simple and linear, but I'd still be interested in the results in the future.)

Still puzzled, though.

[Gregjazz]

Sorry for all the confusion, and if I misled anybody. As I said in my previous post, it does use an impulse. Normally this is generated (as in my SynthEdit project), but since Kontakt can't generate sound from scratch, I had to start out with a small impulse.

[DaveClark]

Hi Greg,

Thanks for the clarification.

Best regards,
Dave Clark