Home >> Markov Constraints >> Unary Constraints

The Applet based on Greedy Markov Constraints exploits finite-Length Markov Processes and constraints to generate controlled melodies in a given style. The applet generates melodies that imitate the style of jazz guitarist John McLaughlin and satisfy two types of control constraints: harmonic constraints and anchor constraints. The harmonic constraints force constrained notes to belong to a specific harmonic context, to create melodies spanning different tonalities. The anchor constraints impose pitches on constrained notes. Changing the constraints it is possible to observe how they control the pitch contour and the harmonic context of the generated melodies. In practice, the training melodies used to build the initial Markov Model and model the John McLaughlin’s style consists in “staircase” scales, transposed in all tonalities. They are simple examples of typical jazz scales, as used in jazz training. The following figure shows parts of the Cmin scale:

For more details see F. Pachet, P. Roy, G. Barbieri. Finite-Length Markov Processes with Constraints in Proceedings of IJCAI’ 2011, July 16-22, 2011, CCIB Barcelona, Spain.



Instructions: how to use the applet

Click on the grid to add or remove anchor constraints.

Click on the chords names above the piano roll to change the harmonic constraints. Each chord button imposes the respective harmonic constraint on the bar below the button.

The button GENERATE generates and plays a new melody that satisfies all the current control constraints. The button REPLAY replays the last generated melody. The button STOP stops the reproduction of a melody.  Each time a melody is played, the notes are displayed on the piano roll. 

The button CLEAR NOTES clears the piano roll.

The button CLEAR CONSTRAINTS removes all the anchor constraints.

The slider on the left controls the applet’s volume.


Related Papers