Relay-Version: version B 2.10 5/3/83; site utzoo.UUCP
Posting-Version: version B 2.10.1 6/24/83; site flairvax.UUCP
Path: utzoo!watmath!clyde!burl!ulysses!mhuxl!houxm!vax135!cornell!uw-beaver!tektronix!hplabs!hpda!fortune!amd!decwrl!flairvax!kissell
From: kissell@flairvax.UUCP
Newsgroups: net.music.classical
Subject: Re: Algorithmic composition: 2 questions
Message-ID: <748@flairvax.UUCP>
Date: Fri, 7-Sep-84 09:40:45 EDT
Article-I.D.: flairvax.748
Posted: Fri Sep  7 09:40:45 1984
Date-Received: Fri, 14-Sep-84 03:48:12 EDT
References: <403@uwvax.ARPA>
Organization: Fairchild AI Lab, Palo Alto, CA
Lines: 30

(tap tap)

I ran a series of algorithmic composition experiments this summer
working on a performance piece.  I used a simulated 1/f noise
generator driven by scrambled digitizations of ambient sound.
The pseudofractal generator was used for all decisions in the
experiments.  I used a variety of tone selection algorithms,
ranging from random chromatic behavior of four voices to rigid
obedience to conventional chording and voicing rules.  Repetition
was constrined by inputs to the composer dictating the total
length of the composed segment and the number of "themes" to
be found within it.  The actual breakdown and degree of repetition
was chosen from a small set of symetries by random selection.

What I learned from all this is what has been observed by others:
The "musical" quality of a piece seems to stem from an interplay
of chaos and order.  The highly random composers produce unlistenable
garbage 90% of the time, but one in ten was very nice to listen to.
The orderly composers produced a smaller percentage of complete
failures, but the successful compositions were less interesting
than the best of the random compositions.  The next step would
seem to be to generate metarules to decide when to break the
basic rules.

Kevin D. Kissell
Fairchild Research Center
Advanced Processor Development
uucp: {ihnp4 decvax}!decwrl!\
                             >flairvax!kissell
    {ucbvax sdcrdcf}!hplabs!/