To hijack the thread a bit, I'm interested in synthesizing sound samples from scratch. Things like footsteps, squeaking doors, wind, birds, percussions and collision sounds. I'd like to have a synth of sorts that can be configured through ~30 parameters to produce wide range of such effects. People often use SFXR and clones  for such purpose. Is there a model that would produce more convincing results while keeping such low number of parameters?
To stay in topic i have synthesized some nature sounds (birds, waters) myself using this synthesizer and a simple two oscillators FM setup with massive pitch modulations for each oscillators, you can hear it on :
And FM squeaking doors : https://youtu.be/25o4Ru8kbpE?t=1404
So my guess is that some FM with pitch modulation and many oscillators may be the way to go.
Percussion can be synthesized with convincing results through resonant models (simple resonant filters with initial noise), this is called modal synthesis, you can find filters parameters for some real instruments here : http://www.csounds.com/manual/html/MiscModalFreq.html
"Fragment is a collaborative cross-platform audiovisual live coding environment with pixels based real-time image-synth approach to sound synthesis, the sound synthesis is powered by pixels data produced on the graphics card by live GLSL code, everything is based on pixels!"
Now, if only there was some way to transpile existing VST/AU plugins into this platform. I've been dreaming of GPU accelerated legacy instruments and FX for a while!