[Smt-talk] Aesthetics of Computer-Generated Music

Marcel de Velde marcel at justintonation.com
Fri Apr 8 07:35:08 PDT 2011

Hello Sandeep,

I'm personally not aware of any computer algorithm that can generate 
compositions that are comparable aesthetically to something like common 
practice period music.
For modern / more atonal music or drone music etc it's a different 
story, but current algorithms have huge problems with a coherent tonal 
side of music.

I think it is also a sign that music theory has not reached the level 
yet that it perhaps could.
My understanding is that if one writes an algorithm (let's say for 
composing a string quartet) on basic rules of counterpoint and 
functional harmony, that the output of this algorithm will still 
generate many unmusical / confusing notes and can not compare to a human 
using those same rules + his/her ears while composing.
On top of that, you will need real humans with good ears playing the 
strings to perform the composition "in tune", as 12 tone equal 
temperament will not sound as good, and current music theory can not 
describe just intonation (only Pythagorean yet this sounds even worse 
than 12tet).

I have been studying just intonation for the past few years.
And I personally think it could make good musical algorithmic 
composition possible. (infact I think it could eventually become a true 
revolution in music composition)
The point is that the algorithms should work the way our ears and brain 
work, use the same language.
And 12 tone equal temperament is definitely not the right basis for this 

I've done a small experiment myself with computer generated composition 
(really small nothing serious, just for research purpose).
It is a true algorithmic composition.
The rules were these:
1) a sequence of completely random permutations of the harmonic series 
up till the 5th harmonic.
2) hold at least one note from the current permutation to the next 
3) i set a limit for the lowest frequency and the highest frequency 
That's it, only a few lines of code, no fine tuning whatsoever.
I would have taken me more code to even simply define 12tet, let alone 
do anything useful with it.

Here's the output:
It's just a random selection, the algorithm goes on forever.
Rendering was done with a sampled choir (Kontakt sampler)
I've since progressed considerably with my just intonation research and 
know now that the output of this algorithm is not correct just 
intonation in many (if not most) parts, and so it also does many 
unmusical things that sound random and confusing (and to put it in 
correct just intonation would mean different and much more complex 
ratios in many places).
But the output is far from random notes in 12tet (which would output 
random 12tone atonal music in comparison).
In a few months I will hopefully begin writing a serious composition 
algorithm based on correct just intonation.
The way it's looking now it's output will be real music, fully emotional 
and coherent.

But in any case..
I think it would be a good idea to use common practice music 
(renaissance / baroque) as the music used for comparison between 
computer and human compositions.
It's music with where the tonal language of the music is clear.
And I think you will find that the computer algorithms have not yet 
mastered this tonal language, and that the human ear / brain is 
currently still needed :)

Marcel de Velde
marcel at justintonation.com
Zwolle, Netherlands

> Dear Collective Wisdom
> could anyone refer me to studies done on computer-generated music as 
> aesthetic entities ?
> I would be especially interested in [double-]blind studies where 
> listeners were asked to rate "meaningfulness"
> or emotional richness of music that was a) either composed with an 
> intent to convey emotion or b) fully generated by computer.
> Of course, the method of delivery (computer players/ human performers/ 
> purely electroacoustic etc.) should be equal for both musics.
> I am aware that, of course, making and listening to computer-generated 
> music is in itself an aesthetic statement, regardless of the emotional 
> import of the music.
> But while this is certainly of interest to me in general, for this 
> particular research I am not interested in that aspect.
> I am interested in what you could call the "emotional reality" of 
> music as it is experienced by listeners.
> Do listeners perceive a "message" in music even if there is no direct 
> human intervention in its "composition"?
> Can listeners make out if something is composed by computer or by a 
> human composer (perhaps using the computer as a tool, or without any 
> computation at all)?
> My hunch is they cannot, and do not really care - but I would like to 
> learn more about it.
> Any hints ?
> Thank you in advance
> Sandeep Bhagwati
> Canada Research Chair in Inter–X Art
> Concordia University Montreal

More information about the Smt-talk mailing list