From winter 2001 to spring 2003 I was researching and developing software and code for projects involved with how multiple computers (2/3 in this case) were able to generate live improvised music and micro-sound music, on their own, as a result of their interactions. I intend to return to these experiments sometime in 2004, but I want a break to digest all these ideas, techniques, failures, successes, blind alleys and the mountain of code I ended up with.
The last stage of this period of the project was the development of a visual programming language (up to now unfinished and perhaps abandoned) as a means to define agents and their system of interaction. The idea was that they would form an autonomous group, a collection of individual members jamming together, by running a set of independent, intelligent sonic agents where no human intervention would take place, once they were started. It was a post-human music performance created by artificially intelligent programs.
Each computer would run a program that would consist of one or more of these independent sonic agents, sometimes using just one computer but running several agents. The computers/agents were networked using Midi, or, if only one computer a local homemade protocol was used, which sent information within the program they contained.
For these agents, I was working on a variety of software architectures and methods of communication. The architecture consisted mostly of perception, objects, feature classified objects, control objects, rule systems, sets of midi generator objects, and protocols of how messages or musical information was shared. Basically it was a Perception-Action cycle, which created feedback loops between each sonic agent. Some systems generated more emergent compositions while others were tightly constrained. During the beginning of the project they responded and interpreted each other’s MIDI note data, musical symbols, but during later stages the information became more abstract and systems related such as states of program variables. The series of projects was based around using MIDI, an extended MIDI technique, which could produce glitch sounds and more electronic sounding textures from the general midi (GM) tone generators into which they were plugged into (Yamaha MU128s). These techniques worked by a continual sending of structured sys-ex data to the synths as a means to continually adjust tone properties and produce gestures of sound. Or using a note based granular synthesis method that produced clouds of sound by using hundreds of notes per second. The latter technique is inspired by an article I read in the Computer Music Journal, or was it Interface?, In the early 1990's. I'll find the reference later.
The project was influenced by the ideas and work of the League Of Automatic Composers, The Hub and the books by Robert Rowe (Machine Musicianhip, Interactive Music Systems), but I wanted to remove the human during performance.
to be continued...

Performances using ISA
I performed 3 concerts using these prilimary systems, all during 2002.

Radar Electronic Sound Bar - Madrid [ click ]
Microsound Lounge - Madrid
Post Digital Tendencies - Madrid [ click ]

Articles about my ISA project

www.ccapitalia.net/paulwebb
El Pais newspaper

MP3 of one system at work during performance
on CCapitalia CRC
click

this is the system I performed at the Post Digital Tendencies in Madrid. The system comprised 2 agents that communicated and were altering the flow of each other’s generative sound code as a result of their interaction. The system was called "codeFlow"

Each agent system comprised multiple generators that could run parallel, based on a technique in which the agent’s state would alter the processing path through the algorithms. I first created the generative algorithms and then expanded these with lots of control statements, which could alter how this algorithm got processed (for example skipping code, repeating code). Sonically I used lots of granular midi techniques, some low chord movements and rumbling sounds and more bassy, glitchy sounds. For some reason the agents during this performance spent quite a while stuck into the same sound type at the start. I guess a positive re-enforcement was occurring, which locked them into each other. I was worried this may last for the whole performance but it did come out.
Listening techniques were based on grain densities, spatial gestures with reciprocations, and frequency domains of grains, levels of changes in which the other could respond with similar or opposite attraction. Then overall compositional properties were being altered. I would need to look into the system again to supply more details.
The comparison of using MIDI to using DSP for generating audio was obvious. The aesthetic quality of glitch at this time was based on DSP technique, which gives it its certain sound quality, which my extended MIDI technique could not reach. But it was really hard work and took lots of investigation to find these sounds with Midi. Compared to my work with digital audio algorithms in which this sound quality just happens, its implicit.