Re: Sarn

Hi,

Regarding David Copes work, I have played
with generating music and text using Markov chains (also
known as a 'Travesty Generator') and using neural net code.

The Markov chain code produced more interesting results, which
had NO elements which were not in the input, whereas the
neural net code could provide answers based on a training set
to an unexpected question ("How would Bach solved having this
crunchy chord inserted into the middle of this particular
fugue", though the program never was healthy enough to ask it
such questions.

Both programs had absolutely no musical intelligence, so that
dropping beats, looping on innapropriate fragments, changing
keys etc, such 'non-musical' things were arbitrary and quite
frequent.

Copes work 'fixes' a lot of the 'non-musical'
elements these 'chew and/or regurgitate' methods have. That
he has used it for making faux-Mozart or whatever is a different
issue, as is whether this is "AI" (which I believe he would say
it is not).

I mentioned him in the context of JdLs tuning
program in that his idea of 'signatures', which tend to give
some of the 'musicality' that the methods I mentioned lack, seems
similar (in some way) to how JdLs program can discover an
'underlying tuning' of a piece. [JdL : Another branch of computer
music research that may be illuminating here are algorithmic
methods of determining what key a piece is in and methods for
determining the meter of a piece].

Bob Valentine