Algorithmic composition/generative algorithms


#1

Hi - I’ve started looking at the literature for algorithmic composition/generative algorithms.

Does anyone have any recommendations for basic reading to start off with ? ie books or papers…

(I’ve got a background in pure maths and non-linear optimisation - many years ago - if that helps guide recommendations)

I’ve been playing with ideas in matlab but really ought to have a look and see what else has been done… been looking at playing with genetic algorithms, markov models and neural nets (as that was my research background before I left academia).

(Long term possibilities are to maybe end up with some fw for o_C, TINRS tuesday or marbles - but to be honest - it’s really just to play with ideas and have some fun in matlab or use it as an excuse to finally learn puredata)

Thanks a lot…


#2

I cant offer any suggestions in terms of literature - but will be very interested in what develops from this forestcaver - I have an ongoing obsessions with creating pieces of music using O&C Quantermain, and specifically using the Integer Sequences - I’m continually blown away by what can be achieved using this approach - Between it and now Marbles I have absolutely no desire to have a traditional sequencer or hook my case up to a DAW.

Make sure you provide updates as you progress, I’ll be keen to try out what you come up with!! :slight_smile:


#3

Me too! (I really like using Grids to drive quanterman). But my reason is that I have no talent :slight_smile:

Will do !


#4

Yeah, there’s the entire OEIS to explore…

I think “algorithmic music” is beginning to split into

a) classical algorithmic composition: humans deliberately design algorithms (usually with stochastic aspects to them) to generate music, based on underlying music theory, or intuition about what might work

and

b) machine learning (ML) approaches (including deep learning) in which humans design a ML algorithm and/or framework or neural network, and the ML then learns the parameters and hyper parameters (and in deep learning, increasing, the architecture is learnt as well) from example data (music). Or the machine just teaches itself, as in GANs (generalised adversarial networks) and now reinforcement learning. I’m kidding, humans are still needed to indicate whether the generated music is any good or not, at least for now.

There a large literature on the former, and a growing literature on the latter. Have a look at stuff like the Google Magenta project, but some searching on Google Scholar is enlightening. There’s a lot going on!


#5

Thanks Tim - I’m really out of touch - but that is how I view it from a very naive perspective (along with deterministic algorithms)
I am much more leaning towards the machine learning side of things (I remember a neural network conference in the 90s where there was a fair bit of discussion about music composition)

I wasn’t aware of the magenta project (thanks !) I’ll have a good look at that. I’m reading a few papers at the moment - I was keen to try to find review papers or a book to give me a leg-up and get up to speed… I’ll keep on searching for review papers for the mo… there seem to be a few publicly available datasets as well - I need to have a good play !

Cheers
Andy


#6

Have you heard of David Cope?

I think his comprehensive approach to algorithmic composition and musical analysis is a great place to start, especially if you comfortable with math already.


https://www.amazon.com/Computer-Models-Musical-Creativity-Press/dp/0262033380/ref=sr_1_1?ie=UTF8&qid=1544224778&sr=8-1&keywords=Computer+Models+of+Musical+Creativity.

I’m not sure why the last link didn’t turn into a nice tidy preview, but it’s a good one also.


#7

Thanks Robert - I’ll check them out!


#8

A lot of really interesting stuff here


#9

Thanks a lot - reading it now!