Have you been on the website recently? They have been working on it for like the past 2 months.
Clearly I have, since I'm suggesting that the data Toady's currently outputting be used to generate actual music that we can physically hear.
I can only speak of what I know (music), but procedural music is very very hard to create. Something you can actually listen, i mean.
It really depends on how conventional and complicated you want it to be.
Writing a computer program to generate twelve tone rows, and layer them in certain ways is not terribly difficult. I wrote a twelve tone fugue by just deciding what features I wanted two different rows to have in relation to each other, and it practically wrote itself.
Writing a program that uses already existing phrases to generate a generic I-IV-V blues song wouldn't be too hard, although it would be predictable. Writing a computer program to generate original music that sounds like a Beethoven symphony would be difficult indeed.
For me, even having a single voice melody with the appropriate tonal and rhythmic qualities play when you come near musicians would just be amazing.
If the program only spits out music with a single voice and percussion, it really would be within reach. If you aren't hard set upon only having a pleasing western sound (which most of the sample descriptions do not describe), then you could easily have it generate a handful of default musical phrases or gestures for a culture or style, that it then overlays.
I have written music that is constructed primarily by layering phrases using a microtonal instrument I built.
Not to get all self promotional, but here are three tracks that use that method of overlaying phrases made with that microtonal stringed instrument.
This one's fairly full sounding with weird samples,
https://soundcloud.com/nethodsod/rhainodaisesAnd this one is much more subtle than most of my music, being meant to evoke wandering desert people:
https://soundcloud.com/nethodsod/arid-porchBoth of those are fairly heavily mixed, but if you don't do any editing at all, you get something like this, which was made with a random tuning:
https://soundcloud.com/nethodsod/nocturne-in-q...
So from all that, I do think that it is feasible to generate a random melody in an arbitrary tuning system, and have the melody fit into an arbitrary rhythmic pattern. Much avant garde music is written in this way, deciding upon available tones and rhythmic qualities prior to any consideration for end product. With only one voice (one note at a time) the music would still sound strange, but notes wouldn't not be clashing with each other.