# Re: Odd questions for you all ;) Degradation algorithms versus Enhancement algorithms

From: Thomas Buckner (tcbevolver@yahoo.com)
Date: Sat Feb 05 2005 - 18:54:11 MST

--- Marc Geddes <marc_geddes@yahoo.co.nz> wrote:

> Let’s do a thought experiment. Define a
> algorithm’ to be any algorithm that takes in
> data
> about something in some sphere and outputs a
> ‘worse
> version’ (as judged by some group of humans on
> average) of the data. An ‘Enhancement
> algorithm’
> would be the opposite – it would be an
> algorithm that
> takes in data about something in some sphere
> and
> outputs a ‘better version’ of the data (as
> judged on
> average by the same reference group of humans)
>
> For instance:
>
> A music degradation algorithm would take in the
> data
> representing any piece of music (listened to by
> some
> large reference group of humans), modify it and
> output
> the modified data which when ‘played back ’ as
> music
> is nearly always judged to be ‘a worse version
> of the
> same musical piece’ than the original (‘worse’
> as
> judged by taking the average of the same group
> of
> humans that listened to the original music).
>
> A music enhancement algorithm would do the
> opposite.
> It would take in the data representing any
> piece of
> music (listened to by some large reference
> group of
> humans) , modify it and and output the modified
> data
> which when ‘played back’ as music is nearly
> always
> judged to be ‘a better version of the same
> musical
> piece’ than the original (‘better’ as judged by
> taking
> the average of the same group of humans that
> listened
> to the original music).
>
> Questions: What would the difference between
> the
> ‘degradation algorithm’ and the ‘enhancement
> algorithm’ be? Would the degradation algorithm
> simply
> be the opposite of the enhancement algorithm?
> How
> many different ‘types’ of musical degradation
> are
> there do you think? (Is there is finite number
> of
> different possible ways in which you could
> given piece of music or an infinite number of
> different possible ways)? Now how many
> different
> ‘types’ of musical enhancement are there? (Is
> there a
> finite number of different possible ways in
> which you
> could ‘enhance’ a given piece of music or an
> infinite
> number of different possible ways)?
>
This is a subject I have thought on for years,
Music has a Kolmogorov complexity like most
everything else; there is finite number of sounds
you could put into three minutes, as there is a
finite number of possible novels (we are assuming
that nobody but transhumans and SAI's want to
billion-page epics, and that's still finite). The
arrangements you would find pleasing to hear are
a small subset, and they have mathematically
interesting properties. It follows that there are
finite types/ways of enhancement, some of which I
will describe.
The music we usually listen to is based on a
twelve-tone scale (even if there are odd noises
and samples thrown in). To go up an octave, you
double the frequency, and this is divided into
twelve equal steps. Twelve is a groovy number
which divides evenly in a lot of ways for such a
small number, and musical theory has been digging
around its divisors and their incestuous
relationships for centuries.

One can imagine getting tired of twelve-tone
been fully explored by jazz musicians early in
the last century and that more recent popular
music was simply regurgitating what they had
discovered; I was scandalized, but this view is
not necessarily wrong, and Brian Eno says that
newer music innovates in other areas while using
older forms (i.e. sonic textures and samples,
laid over the same twelve tone scales). Other
microtonal scales exist which divide the octave
in many different ways (quarter tones, 43-note
scales, 31 notes, and so on). Google 'microtonal
music' for more. This sort of music has been
around for quite a while too, but it's an
acquired taste limited to ethnic minorities and
scholars, and perhaps needs its own Mozart or
Beatles to reach the masses. Microtonal scales,
however, theoretically offer many forms of beauty
which have never been heard by the human ear.

It is worth noting that what humans find pleasing
has mathematical interest as well, and
corollaries exist in other arts; a Discover
Magazine article shows how Jackson Pollock
masterpieces have fractal dimension, while fakes
do not. The best music contains interesting
melodic/harmonic patterns, interesting rhythmic
patterns (time broken up in mathematically
interesting ways), and, if there are lyrics,
interesting word structures and emotive
components.
For example, one book recommends that songwriters
sit down with their lyrics and lots of colored
markers. Rhymes should be marked with one color,
internal rhymes another. Metaphors, similes,
onomatopoeia, visual words, words which evoke
touch, smell, etc., all get different colors.
Ideally, the lyric sheet will be a riot of color
when you are done, giving a visual key to the
complexity of the lyric. I recently read a poet
who explained that a poem was like a machine
where every single word should be necessary, like
a screw or a cam.
In addition, the melody should reflect the lyric,
and vice versa, as for example in "Animals" when
David Byrne sings "down, down, down, down" in
descending, gruff notes suggesting a command
given to an unruly dog.

If you have followed me this far, you can see
that there are layers of psychoacoustic
complexity underlying the best music. In this
case, a 'degradation algorithm' is easy to make,
because almost any random change you can make is
which can in reality be a fiendishly subtle and
balanced creation. An 'enhancement algorithm', as
you can see, is likely to be limited in its
reach. It may be effective on a very simple or
flawed piece of music, but music of high quality
is already the product of a number of enhancement
algorithms which the artist has consciously or
subconsciously applied (the latter case will be
described as 'having a good ear' just as Jackson
Pollock's ability to create fractal dimension
with drips and splatters might be termed 'a good
eye').

Tom Buckner

__________________________________
Do you Yahoo!?