Recently, I have become very interested in machine learning and A.I. in musical production.

So, in the spare time, I started a machine learning project focused on music creation, using my own composition style.
Using Magenta [1] (a python library based on Tensorflow) I trained an A.I. with all music i composed during the last twenty years (almost all, a lot of songs was never transcripted).

The result was (obviously) a series of strange melodies: nothing usable without a large amount of adjustments, but a good starting point for a new composition.

From this experiment, I've got some interesting points, that i turned into a collection of 4 strange, experimental tracks.

Unusual, but not to different from my recent compositions.

You can listen it on Spotify:


  1. magenta/magenta: Magenta: Music and Art Generation with Machine Intelligence