Re: [Faudiostream-devel] FAUST + physical modelling

[Stephen S. <rad...@gm...>]

On Sun, Nov 8, 2009 at 6:45 AM, Stefano D'Angelo <zan...@gm...> wrote:
> Hello,
>
> First of all I want to thank you for this fantastic piece of software
> that is FAUST. :-)
>
> I am writing my master thesis on physical-based modelling techniques
> for sound processing and synthesis and the objective of my final
> project is to implement a system to develop a system to specify and
> deploy physical system simulators.

Great!  I did something similar, I have been doing work on haptic
interaction with physical models, and to help with this I translated
one of the STK models (Bowed) into FAUST and achieved quite a
significant speed-up for my project using a TI DSP.  I had been
meaning to post about it sometime..

I found FAUST a nice architecture for this because it made modifying
and re-parametrizing variables, experimenting with changes, etc., much
easier.  I intended to eventually hook into FAUST's GUI architecture
for communication with the DSP.  I _loved_ how it could be used to
generate block diagrams that pretty much directly express my intent,
although sometimes I wish I could give the compiler "hints" on which
blocks to make visible.

ANyways, yes, I think FAUST is pretty fun for physical models.


> Now, I already wrote a little subsection about FAUST in my thesis ;-),
> but I am also considering whether it would be possible to use it as
> the code generator of my system... it could work like this:
>
> * a bunch of models of physical components are written in FAUST, and
> maybe carry with them some additional files needed for physical
> modelling;
> * a higher level language specifies the topology of such components in
> a patch - the patch in such language is translated to FAUST by an
> ad-hoc program, and is also reusable as a model in other patches;
> * FAUST works as usual, no changes needed.
>
> If this works well, I am doing all the work myself, so don't worry ;-)

Sounds like it could be done by preprocessing some files and
generating a simple 'process' to glue them together.  It would be cool
if the LLVM back-end could be used to generate and execute that model
in-memory.  Is the LLVM port working?


> Anyway, the suitability of FAUST should be pretty much dependent on
> how FAUST actually "merges" the blocks together... for example, if an
> input signal for a block is a known quantity/function/output from
> another block, will the operations involing that input be simplified
> in the generated code? Or, in other words, will the blocks be really
> "merged" together, or are they just going to be run one after the
> other?

I think the FAUST compiler is pretty good at optimizing whatever
combination of blocks you throw at it.


> Also, it would be cool to have a LV2 backend... if it's an easy enough
> task, I can probably do that. And it would be even cooler if there is
> some kind of FAUST interpreter, but that is probably too much for me
> to do. :-)

I'm not sure of the advantage of an interpreter, but just having some
debugging helpers might go a long way.  Sometimes FAUST's error output
can be pretty hard to read.


Steve