Calling all engineers!

Is there anyone who can get me started on my way to physical modeling? I would just like someone to provide in laymen's terms the basic conceptual difference between creating a software instrument by modeling it vsd FM synthesis.

Or is FM synthesis + studying spectrograms of instrument samples and using bp filters, noise bursts, etc. to match them the extent of it? (I've tried it so far to little success).

A generic answer is desirable, but just so you know I am going to attempt to use PD to make acoustic piano sounds as a first project. Maybe someday I can have a whole software orchestra do Mahler, why not?

P.S. If anyone has not looked into Yamaha's AN, here is what inspired the project:

http://www.vintagesynth.com/yamaha/ex5.php

You probably want to look into the Karplus-Strong algorithm;
I suspect that getting it to sound more like a piano than a synth will lie in getting the inharmonicities (sic?) right; I'm curious if anyone else here has looked into that.

That AN doesn't really look like it's what is typically referred to as physical modelling. It looks like it's just a digital emulation of analog synthesis, which isn't all that special.

Physical modelling is really what the VL part of that synth is. It uses computational algorithms to model the propagation of energy through a given body based on our current understanding of acoustics and the physics of the body. There are several approaches, but probably the most common uses digital waveguides, which are basically just delay lines that represent acoustic energy going back and forth (e.g. along a string). The difference between making it "just a delay line" and refining it so that it models real world instruments can get pretty complex. It can be pretty daunting for a first Pd project.

It sounds to me, though, that you are thinking of it more in terms of listening to a sound, synthesizing the individual elements of a sound, and then adding them together to create the overall sound. That's a whole other ball game and likely a bit easier. The thing about that approach is that you can pretty much mix and match any type of synthesis you think is needed, whereas FM synthesis is just FM synthesis. As an example, you could try some form of FM to model the resonances of a cymbal and mix in some filtered noise give it some shimmer and make it sound less clangy. That uses FM and subtractive synthesis.

You're exactly right. Thanks for the concise explanation, Maelstorm. It's actually not a first project, just the first one where I am experimenting with a precise idea of the sound I want to end up with.

It sounds very complex, but would you say more? What kind of homework is involved in this problem, assuming a knowledge of physics and calculus? (It doesn't matter to me if it takes some considerable research.)

found that :

http://puredata.hurleur.com/sujet-3751-physical-modeling

This is probably the most comprehensive information on physical modelling out there, focusing mostly on digital waveguides:

https://ccrma.stanford.edu/~jos/pasp/

Thanks nau, I searched the forums first but I hadn't encountered that one.

Maelstorm - great link, thank you. Comprehensive and just what I was looking for . . . DSP heaven!

wow, cheers for that link, Maelstorm; looks pretty cool and complete indeed!

@mapadofu said:

You probably want to look into the Karplus-Strong algorithm;
I suspect that getting it to sound more like a piano than a synth will lie in getting the inharmonicities (sic?) right; I'm curious if anyone else here has looked into that.

The inharmonicity (which can be modeled in a simple way by inserting a first order allpass filter in the feedback loop) is part of it, but I think the stimulus (hammer hitting string: this can be sampled or synthesized) and striking position (feed forward comb filter) are very important. The soundboard and string coupling are important too, but those are more complicated to model. The soundboard can again be sampled and convolved with the hammer impact.

It's not that difficult to just mess around with these basic elements and get something that vaguely resembles an "acoustic something or other". But a decent piano isn't particularly easy.

Just keep in mind, theoretically there is no big difference between a guitar and a piano: "Hitting strings". But since the two don't sound very comparable (even starting with the sound-envelopes! & even if you hit your guitar with an hammer - I mean only the strings..) the rather subtile changes must be very important...
A piano can generally be seen as a "hammered guitar-deluxe" with several coupled strings per tone (and I guess the tone-tone-coupling is higher (not neglectable) as well). The rest should just be a different filter, that can be convolved with the hammer-input, as it was mentioned..

I have no idea, but I feel like the inharmonics, so allpass-filters, are not that important. A Piano doesn't sound very bellish to me..

I don't mean to be argumentative, Flipp, but there's definitely a big difference. If they don't sound the same (which they don't) then they are different. Granted, they both use the basic model of energy moving along a string, but the strings are different (much thicker and longer on a piano), they are excited differently, the resonating body is significantly different, pianos have three strings per note, there are much more strings that could be sympathetically vibrating on a piano, there are different places to pluck the string on a guitar, the same note can be played more than one way on a guitar (and they each sound a little different), etc.

Individually, these things may seem minor implementation-wise, but they all add up and matter in the end. Especially if you want to make your model expressive, which is what really make physical modelling so interesting. You can use control models based on real physical interaction, which other synthesis models don't intuitively allow.

If I got you right, isn't that what I basically said..?!
With the difference that I physically categorize a piano to be rather a string-instrument than anything else. Without doubt a piano is a piano and a guitar is not...
Maybe think of this way: Guitar -> Harp -> all those old instruments, something inbetween plucked and hammered -> Piano.
I mean 1 or 3 strings - whatever.. (upper keys usually just got 2!). And the higher the notes the more identical the instruments sound...
Anyways think of trumpets or even organs, although a key-instrument, -they are far more different.. Bells, Xylophones etc. Opend ends, closed ends, dispersion and so on.
The one or two things that really differ are the body - the "loudspeaker", and the coupling between soundgenerators. They are fundamentally different, as opposed to the hammering - a guitar can be hammered as well, and a piano can be plucked ..or even bowed.... An organ or trumpet can't be! But these eg. can be overblown..

Sorry, I may not have quite understood what you wrote. It just seemed to be an oversimplification to say that there's no big difference between a piano and a guitar when the only thing they really share in common is vibrating strings.

i learned a lot from tinkering with Karplus-Strong, it might be as entry-level into modeling as you can get but it can be very fun to tweak with various filters at different places in the patch. highly recommend building your own K-S patch from scratch.