The following discussion took place via E-Mail between
Jim Dorsch and Doctor Nerve's Nick Didkovsky.

>> First, do you have any background (besides your paper, which I have) on
>> computer-generated music? Who pioneered the field?

Extracting from Charles Dodge and Thomas Jerse's Book "Computer Music, Synthesis, Composition and Performance" (Schirmer Books), automatic composition was pioneereed in the early 1950's. Dodge and Jerse break the systems down into two general categories: 1) Aleatoric or stochastic music (statistical characterizations of a random process), and 2) deterministic systems which might use a catalog of compositional processes in various permutations to produce musical material.

They further break down stochastic systems as those which use independent outcomes and those whose outcomes depend on previous outcomes. (like dice whose each throw has no memory of previous throws, versus markov chains where outcomes depend probabilistically on what came before).

Harry Olson used Markov processes to create computer music statistically similar to the music of Stephen Foster by analyzing transitions in the original compositions, and from these tables, generating new works. This was in the 50's.

Another method, the "generate and test method" used random numbers to suggest possibilities which another part of the program judges for acceptability. This was used by Hiller and Isaacson in their famous "Illiac Suite". Simple example: to create a random melody in the key of C, generate a random note and throw it out if it's not in the key of C.

Xenakis has done very important work in stochastically generated music.

James Tenney made some radical breakthroughs in the 60's during his time at Bell Labs (check out the CD of his work on Artifact Records, available through Frog Peak Music, Box 1052, Lebanon, NH 03766

>> How is HMSL structured? How many random variables come into play, and what do
>> they define? As the programmer, you must have an influence on the outcome.
>> Does the program follow rules so that the instruments behave together in
>> certain ways? What do the musicians think of having to play the technically
>> challenging passages?

HMSL, or Hierarchical Music Specification Language is a programming environment based on the FORTH programming language. It runs on Macs and Amigas. I use the Amiga version, which is written in Phil Burk's and Mike Haas's "JForth". Conceptually, there is a layer of object-oriented development tools (called ODE) that sits on top of FORTH. On top of that layer sits HMSL proper, which includes the time scheduler, music-specific tools and objects, etc etc etc.

You can write an HMSL program using all of these layers, or only one or two. That is, you can write major applications that look nothing at all like HMSL, because you have JForth sitting there, which is a full Amiga programming system. Or you write a lot new classes of cool objects that do object-oriented things and thus write minimally in JForth itself, building a musical hierarchy out of these new and existing classes of objects. Since HMSL by itself doesn't do anything, you can write pretty much anything you want, including music systems that use no random variables whatsoever, or systems that are themselves created randomly using as much memory as your machine has. (Rosenboom has done a lot of work in randomly structured hierarchies). If you choose to use random elements, they can define anything you program them to define - it's not a set thing. There are no limits except for hardware. An 8 meg computer could have millions of random variables determining anything from the color of the computer's screen to the timbre of an audio wave or the average speed of the melody which stochastically morphs into a backwards version of itself. So, as a programmer, you have to work harder, because HMSL isn't like "Jam Factory" or some similar music software that is built on a particular paradigm, and runs within predetermined bounds.

As a programmer, you have total control over implementing not just a new piece, but a completely different way of thinking about music - a theory - programming the theory and seeing what comes out of it. It's deep! HMSL was designed to follow and impose as few rules as possible. It is very good at being non-stylistically biased. HMSL composers sound nothing at all like each other (check out "Hallways, 11 musicians and HMSL" [CD], also available from Frog Peak). Of course, as a programmer, you can write as many rules as you want to, and build a music system out of that - like an expert system perhaps that solves msuical "problems" by applying a rule-base. But HMSL doesn't do that by itself, nor does it persuade you to try that, or not to try it.

Doctor Nerve has gotten really good at playing difficult music! They're into it, as am I. I don't write in ways that are easy for a particular instrument, but since I write stuff that's really hard for guitar and have to sweat and work it out for myself, it's not like I'm getting a free ride by writing parts that are easy for guitar while the rest of the band's killing themselves. We're all killing ourselves.

>> How many runs typically take place before you get a piece of music you like?
>> Are all NerveWare tunes from the same program? I believe you wrote a new
>> version. Is this just a refinement? How much of a tune comes from you and how
>> much from the computer? How do you avoid a sterile sound?

With the first version of the Doctor Nerve automatic compsition software and the current version, the ratio might be something like 5:1. that is, I might hear four performances of software-generated raw material before I hear one that has a germ of something really cool in it. All the NerveWare tunes are from the first version, called DrNerve.hmsl. The new version is really a completely different system. I sent you info about it. It uses parameter curves to build musical behavior in broad brushstrokes, and fills them in stochastically with playable material. It has lead to very different results, and has demonstrated itself to be more flexible than the first program. I have modifies the new software to write for three different ensembles: Nerve, The Bang On A Can AllStars, and The Crosstown Ensemble. The latter two were commissioned pieces which premiered last May and last March respectively. The piece for the Bang On A Can Festival is entitled "Amalia's Secret", and is being performed by the Allstars coast to coast and on their European tour. The piece for Crosstown is for chamber orchestra and includes a string quartet which I am really happy with. So, the software doesn't seem to tend toward a particular instrumental configuration, which is very satisfying philosophically. In fact, the core of the software was developed for the piano music CD I produced on Pogus Productions (PO Box 150022, Van Brunt Station, Brooklyn, NY 11215-0022), called Flies In The Face Of Logic. It's piano music by myself, Steve MacLean (an early Nerve member), and CW Vrtacek. Mine piano music on that cd is generated by an early prototype of the software which was later adapted to become the Nerve/Crosstown/Bang software.

I avoid a sterile sound because the musicians are such monsters and I like to interpret music aggressively. I once heard a muzak version of Day Tripper, for instance, which was utterly repulsive and schmaltzy. That's an ensemble's interpretation of a piece, and shows how radically different an interpretation can be from the intentions of the composer, or in the case of our discussion, how personal, sweaty, and intense a piece can be interpreted regardless of its origins.

Computer-generated music does not have to sound anything like our science fiction ideas of computers (HAL singing Daisy in 2001, or SciFi Robots bleeping videogame melodies), and Nerve has proved this over and over again. It's stunning to me that so little computer-generated music has a radical, aggressive edge to it. I think that's because so much of the history of computer music was written in academic surroundings, due to the cost of the equipment, so the aesthetic historically reflects the status quo of those hallowed academic halls. Now that any lunatic can buy a computer, you'd imagine that more radical things would be happening, but that doesn't seem to be the case.

Computers are usually reduced to the musical equivalent of word processors - contributing nothing of what they do uniquely to the work, any more than switching from pencil to pen would affect the creative process of writing a short story. The reason is that it's hard to mine the treasures of writing your own software. It's easier to adopt someone else's vision of what computer music is by buying their product. So computer music in the PC explosion becomes another opportunity for "virtuoso consumerism", as Ron Kuivala would say; buying products instead of creating your own.

Computer music people I really like: The Hub, Chris Brown, John Bischoff, Tim Perkis, Mark Trayle, Larry Polansky, Xenakis, George Lewis, James Tenney, Godfried Willem-Raes, Judy Klein, Phil Burk, Robert Marsanyi...

Back to Doctor Nerve Main Menu.