Music for Humans by Machine

Music generation from statistical models

This research explores the generation of new music for live musicians, based upon a corpus of existing music.

Flow My April Joy : Dowland in Dowland / Metheny in Dowland (2016)

A collaboration with Darrell Conklin, this work analyses John Dowland’s  (1563-1626) Lachrimae, or Seaven Teares (1604), seven pieces for viol consort based upon his more famous Flow My Tears (1596):

The machine analysis extracts phrase, harmonic, melodic, and rhythmic information for a generative system, which uses the first pavane, Lachrimæ Antiquae (Old tears) as a template. The template provides the following information:

  • first and last chord in each phrase
  • last note in each phrase
  • length of each phrase
  • rhythmic (onset) placement of each note

The original Dowland was comprised of five parts: cantus (melody), altus, tenor, quintus, bassus. The five parts were generated, and set for recorder and string quartet.

The following work is an example of style-mixing. Dowland’s Lachrimæ Antiquae (Old tears) remained the template, but the data for generation was derived from the music of Pat Metheny; specifically the tunes FamilyFarmer’s TrustFollow MeGrampa’s GhostLakesSometimes I SeeStranger In TownThe Bat.