Damnit! I had typed out a long response and it got ate up in some kind of server crash. Grrr....

Anyway, I liked what I typed, so I'll try to quickly rehash it.

So, I have in mind two innovations: PGA (procedurally generated audio) and "sound language."

First off, I'd like to forget the idea of procedurally generated music. I don't think this will ever evolve beyond either very mechanical sounding pieces or the type of "ambient" music with no melody or tune. There will never be a substitute for a composer, and even one of basic skill will likely surpass even the best computer-made music. The problem is that music, like good writing and so on, relies too heavily on seemingly arbitrary choices made by a composer, whose grasp of the whole goes far beyond what is possible in an algorithm. At best, a computer could only mimic or imitate, but I doubt anyone but the most tone deaf person could tell at once it was made by a computer, not a composer. I'm not an expert musician by any means, but I know enough to know it's more than a computer can handle. There may be new kinds of music that a computer could do well, and perhaps the "composer" in that sense would be the one shaping the algorithms. Still, it seems like a dead end to me.

Conversely, computers would seem ideal for PGA for sound effects. It seems easy enough to me to record a few thousand different gunshots, have the computer analyze and data mine, and come out with some type of algorithms that could generate a unique gunshot on-the-fly based on the parameters of the virtual environment (compensating for distance, acoustics of the room, etc.). Screams and laughter would seem more daunting challenges, but again I see no reason why it couldn't be done given a large enough pool of data for the computer to mine. I'm particularly interested in bird calls and nature sounds, since there seems to be some real potential to use data mining to arrive at believable "alien" soundscapes just by blending in all the different sounds from many different zones. I guess you'd need to limit yourself to climates and zones that are roughly equivalent to the virtual ones (i.e., a forest rather than a jungle or desert). There seems to be a point where few people outside of experts care, though--I mean, how often have you watched a movie and heard a bird and thought, "That's impossible. That bird doesn't live in that part of the country" or some such. Then again, since we're just dealing with data and algorithms anyway, I don't see why it would be THAT difficult just to get it right. How hard is it just to look at a book and find out what birds and such you're likely to hear? But, I digress.

So, if you wanted to make a PGA of a particular bird call, just get recordings of a few hundred or thousand such calls from different birds, have the computer analyze the data and come back with algorithms with the correct parameters (i.e., what is possible given the range you recorded), and viola! I realize I'm making it sound trivial whereas I'm sure doing the data gathering and analysis is serious stuff...But, you get the idea.

Now, sound languages. What I have in mind here is the little sounds we're used to in games, such as the "jumping sound," plopping on tiles, sounds that play when you get coins and such...Okay, now I'd say that this stuff, when it's shared across enough games to become conventional or standardized, is the rudiments of a very simple language. My idea is simply to take this concept much, much further, in effect reversing what I said about PGA above. Instead of having the computer learn the sounds of our world, we have the computer teach us the sounds of its world. Given enough time and some PGA techniques, I'm sure we could end up with some very sophisticated and elaborate sound languages that could add a whole new dimension to gameplay, particularly since you can do things with games that you can't with other media. I have a rough time even imagining what this could sound like, but perhaps games like Captain Blood offer a glimpse. Indeed, I was thinking about speech synthesis as well. Again, I don't think we will ever have speech synthesis that is indistinguishable from a real person. But why bother? Instead, we could learn to understand the speech the computer is capable of synthesizing. So, just have a game set in a world of robots where we expect the robots to talk like that. Go with what works.