Sarn, Cope, and Markov

Sarn,

Your recent post on different types of excursions into algorithmic
music making reminded me of experiments I did a few years back with
Markov chains (AKA Travesty generators) and Neural Nets.

The Markov chain music generator I had sould input a text based
score, and some of the functions it supported were 'addition',,
whereby a new source was added to an existing data base, and
'invert probabilities'. The first function would create bridges
between different musical sources. The second would start
emphasising the statistically unusual aspects rather than usual.
This type of tool is easy to build and generated some interesting
mslodies.

The problem with that tool is that a query has to find an exact match
to get a result. For instance, a Bach database, faced with 12 notes
from Webern would not generate an 'answer statement' along the lines of
how Back woudl finish those notes.

Neural Networks have no such limitations. Unfortunately, the ones
I worked with got into loops too easilly.

Just some thoughts and no, I lost the source code.

Bob Valentine