This page provides complementary information to the paper “Markov Constraints: Steerable Generation of Markovian Sequences” by F. Pachet and P. Roy.

Markov Constraints in action: the Yamanote demo

In the Yamanote demo (video on the right) you can see a simple application of Markov Constraints. By selecting a corpus (in this case, the Prelude in C by Bach and the melodies of the Yamanote metro line, in Tokyo) we can generate new melodies. With this technique, we can apply arbitrary constraints, generate the best (most probable) solution and remain in the style of the corpus itself.

yamanote demo


Home >> Markov Constraints >> General Markov Constraints

Yamanote Demo

Generating new melodies in the same style of the Bach's Prelude and of the Yamanote metro line.


Pierre Boulez

Virtuoso in action

This video shows the system in action with the Sony Move Controllers.






The Boulez Blues example

The Boulez Blues is a blues chord sequence which is the most probable Blues, using Charlie Parker's chord transitions Markov model, that ALSO satisfies the AllDiff constraint (hence its title):

C7 / Fmin
Bb7 / Ebmin
Ab7 / Db7
Dbmin / Cmin
F7 / Bbmin
Eb7 / Abmin
Gmin / Gbmin
B7 / Gb7
Bmin / E7
Amin / D7
Emin / A7
Dmin / G7

The training database for the 12 bar Blues sequence contains the 22 Blues from Charlie Parker’s Omnibook, expanded to 24 chord sequences (2 per measure). They are here translated in C. Only three chord types are considered (7th, minor and half diminished). When sequences are translated in all keys, domain size for each chord variable is 36 (12 pitch classes * 3 chord types).

This applet illustrates an interactive melody generation example, showing the algorithm work on a jazz improvisation problem, inspired by Al di Meola.

The Virtuoso project makes use of Markov constraints for the generation of virtuoso jazz melodies.
Check the videos on the right for watching the Virtuoso in action.

Related Papers