[sdiy] Piano physical modeling

[Richard Moore]

Hi guys,

This may be a bit off-topic; it's about physical modeling-based synthesis.  Aaaannnd it's pretty long.

I love the piano and I'm intrigued by piano preparations and extended techniques, but they're often awkward/difficult/destructive to use.  I would love to be able to quickly change hammer materials/positions, or apply mutes at harmonics or near the bridge, or simulate driving the strings with magnets and an external signal, etc. 

Instead of building lots of ridiculous modifications to a piano (like this one I made a several years ago: http://oprahsfavoritedeathmetal.blogspot.com/2013/10/poor-old-piano.html), I've been thinking about the possibility of starting with a good strings/hammer/soundboard model and modeling the other desired interactions with the strings.  I don't know what an effective interface for such an instrument would be, but I'm thinking of a midi keyboard with a string or ribbon controller that could be used to indicate where and how much (non-hammer) force should be applied to the string models.

Directly filtering/waveshaping piano samples (without modeling) would be significantly easier, but I don't think that's what I'm looking for.

My questions are:

1. Is there a nice physical modeling synth that will already do things like this for me? 

2. Is there a good mathematical model of the piano that I should check out?  The one I've been playing with so far is here: http://www.music.mcgill.ca/~harry/pdfs/Ch_Saitis_MA_Thesis.pdf

3. Any insights to modeling physical interaction with the strings?

4. What sort of performance interface would allow the forces on or response of a string to be measured for use in such an instrument (ribbon controller, noise-driven hand-manipulated reverb springs, single guitar string with a piezo under the bridge)?

Here's my partial MATLAB implementation of the piano model that Saitis presents:

http://www.mathworks.com/matlabcentral/fileexchange/46984-saitispianomodel-m

The hammer force is not yet implemented (now I'm just using an initial condition shape).  So far the sounds I'm getting from the string model are pretty underwhelming.

Thanks,
Richard Moore