Dear reader,

after the experiments with Bayesian inference and pure coordination, I decided to investigate the possible intersections between Finite State Machines and Meta-Games for Free Musical Improvisation.

Historically, Finite State Machines have been brought into the domain of music via the almost exclusive path of algorithmic composition and/or generative music. Most notably, the work of David Cope (Cope, 1991, 2000) is at the forefront of this area. Cope is the creator of EMI (Experiments in Musical Intelligence) and he has investigated thoroughly the vast possibilities of Markov Chains and FSM at the service of computer-assisted composition.

I instead followed a different route, where the computer is used merely for interfacing players and generating the transitions between states as well as forwarding the subsequent information to the improvisers. They, in turn, have the possibility to veto such transitions (for example if they like what is happening at the time) or request new ones, at discretion.

Here the states of the machine are musical scenarios.

The machine consists of seven distinct states, whose transition probabilities are determined in advance.

These states represent the actions/strategies that the improvisers are supposed to realise during that particular state. For moving between states, a type of transition is stochastically chosen amongst seven possible choices. Furthermore, an archetypal interaction (seven different types) is also associated with the current transition and/or new state (= new musical scenario).

The above elements of the machine are named, respectively, Linear Function, Transition Function and Relational Function. When a state self-transitions, no change is output in the other two categories.

Those of you familiar with T. Nunn’s “The Wisdom of the Impulse” (2004), will recognise the above functions.

The actual model is based on what is known as the ‘Birth-Death Process’. This is a particular case of continuous-time Markov process, where all states are connected sequentially, and where the states can only transition forward, with probability p, or backward with probability q.

This particular implementation, however, also introduces a self-transition for every stage. Therefore, the probability of any given state to self-transition, will be given by (1- p-q), for that particular state.

Here is a pic:

I will keep this blog brief and spare you the maths and the blah blah…

There is also a bonus material section, that players can request at will. This features graphic scores and a 1/f Voss note generator. For reference to its workings, please see

Still loads of improvements to make, this is only a prototype…

And this is an instance of it: