Four Short Accidents


These four short movements were entirely generated by a deep neural network, which I had in fact built to serve another purpose; the AI was supposed to learn to model 3-6 part vocal music of Palestrina, a task on which it clearly performed very poorly.

That said, albeit unintentionally, the output was still very interesting to me. It generated movements with a consistent form: an opening quote from Palestrina's corpus before cascading chordal phrases, pausing on a few long, sustained chords. There may follow more cascades and pauses, before the movement converges on a single note.

The model demonstrates some understanding of voice leading logic, albeit in a style far more contemporary than Palestrina's, and how to highlight and move between certain focal pitches and gestures, which gives shape to the music.