<long article by John Walsh omitted>
John contended that tunes are so rare among the possibilities
that any scheme will work.
I contend that in that huge space, tunes are very, ..., very
tightly grouped in one "corner" of the space. Most of
the possible tunes ("most" being one of those mathematical
understatements that are so huge that one runs out of
words) would sound like random notes. If in fact you
take the output of a random number generator and feed
it into a synthesiser it just sounds horrible (like Schnitke
or Schoenberg only worse). No real tune is like that.
The chance of producing a listenable-to tune from
random numbers is pretty much negligibly small.
Efforts at computer composition *do* use random numbers
but not to directly determine note pitch. There have to be
lots of higher order constructs to define pattern. To put
it another way - real tunes have low entropy.
As an illustration of this, many years ago I worked in
IBM where they insisted that I forget my password
every month (if you get my drift) and furthermore
that each month's new password had to be dissimilar
to the last 16 previous ones etc.
I had written a synthesiser program for an Amiga that mapped
the keyboard into notes (a sort of cross between a melodeon
and a button accordion) and to generate passwords I thought
of a tune and just "played it". So the Irish Washer Woman
was rjhffbffhfhrkj and so on. All was fine for a few months,
then it started complaining that most of my new passwords
were too similar to some old password.
If you like "searching for tunes is hard, because to
a first approximation all tunes are the same".
Regarding John's conclusions
1. The conclusion might be right, but I think the argument
is wrong. What the various algorithms do is to project
the tune space which has a very high dimensionality onto
a much smaller space. We then hope that "similar" tunes
land up close and "dissimilar" tunes don't. For instance
a hash code would *not* work here but many things
will. What matters is the distribution in the projection
space, not in the original.
2. Yes, probably good to have several algorithms.
3. Experiment - yes!
Laurie
To subscribe/unsubscribe, point your browser to: http://www.tullochgorm.com/lists.html
