DavidOrr 2014-09-18 15:21:38 (Updated 2014-09-20 22:42:54)

It does indeed have a name! Though I'm not sure it has become the "industry standard" term yet. I'm gonna stick with it.

As far as getting your feet wet, here is a speed-run through writing generative audio.

----------------------
Preface: If you've ever written and rendered musical stems (for film, animation, Royalty-free sites, libraries) you've essentially created generative music. If you haven't;

1. Write a piece of music. Don't worry about any of this generative music stuff.
2. When you're done, play around with muting select tracks, so you are only hearing select parts of the music at once.
3. Find a combination of parts you like on their own (maybe percussion and bass?), and render them.
4. Now render the remainder, WITHOUT the instruments in the first render.
5. You know have 2 stems. Open up a new project file in your DAW of choice, and place them on two separate audio tracks. Don't worry about tempo syncing -- just line them up so they begin at the same place.
6. Play the piece, and experiment fading between the two tracks -- have one track silent while the other is playing, and then switch. Automate fades in random places, and just have fun experimenting.
7. Congratulations -- you've just emulated a AAA generative music system in your DAW!
----------------------

Perhaps the tutorial was unnecessary (and AAA systems are more complex than that), but the concept really is that simple. I've been experimenting with 2-4 different "stems" (portions of the full track), and seeing the results I've gotten. There is a very different compositional process to writing generative music (perhaps something I'll write about in the future), but this is a good way to get started. :)

DavidOrr 2014-09-24 13:26:38 (Updated 2014-09-24 21:42:57)

The easiest approach (and the one used by most in the industry), is to forego loops altogether. You essentially have a 3 or 4 minute track split up into multiple parts, and each part is triggered on/off based on a set of pre-determined conditions. This is called "adaptive music", which I consider a form of generative music. This works pretty well, but (as you suggest) it can be rough on thematic material. Fading a theme in half-way through its statement is not ideal. Luckily for the western market, composers have moved away from thematic writing in favor of a more cinematic underscore style. This works perfectly for this type of generative audio.

With that said, I've been experimenting with having shorter music sections (30 - 40 seconds) that can be pieced together based on a set of music mapping conditions. For instance, take hypothetical section A, B, C, D. Once a section plays, it can move onto a new section of music:
A: can proceed to B, C
B: can proceed to A, D
C: can proceed to A: B
D: can proceed to any section

Each section has multiple layers ("stems"), which can be faded based on game parameters set by the developer.

With a system like this, you can keep the integrity of themes, because you can program the system to play the thematic material of a section in its entirety. Essentially, the system then "knows" where the themes are located, as it has a skeletal framework of the music (defined by the sections). Furthermore, you can now write in modulations (that aren't the same every time), and add an additional level of variability to the soundtrack. If a section ends in a different key than it began, just have it move to a new section in an appropriate key. No need to stick with D minor! :)

There's a lot there, and it really needs to be elaborated on to make full sense. I'll make an in-depth post in the future about my approach to generative audio. Krin and I are working very closely on bringing something new to the generative audio field (what I shamelessly refer to as "next-gen generative audio"). Most of the games we release now use some form of generative audio, and each iteration is getting more polished and diverse.

DavidOrr 2014-10-03 16:10:13

Using the term "Generative" vs "Adaptive" is more semantics than anything else -- I recently updated the post to attempt to clarify a bit. Currently, a lot of people in industry are referring to what I describe as adaptive music, with the implication that generative music is more centered around note generation.

Personally, I see adaptive music (as seen in some modern games) as a first-step towards truly generative music. Generative music can be determined by a set of rules, and there is no reason why groups of notes cannot be treated as a single event in the generation process. The generative aspect, therefore, is not within the notes themselves, but within the combination of layers in the music.

Again, this is purely semantics, and I'm not sure yet if it is worth going against the terminology of the industry on those grounds. I feel I can make a compelling case for using "generative" over "adaptive" -- but at the end of the day, does it really matter?

Thanks for your input!