A Distributed Representation for Chords (Section 2.2)
Chord progressions sampled from the tree model
Chord
progressions have been sampled from the tree model in Section 2.2, with
20 possible states in Level 1 and 36 possible states in Level 2. The
parameters of the model have been learned on the training set using
the EM algorithm. Arpegiated versions of the chord progressions
(playing each note one after another instead of playing them all at
the same time) have been included to show that (more) interesting
music can be created from sampled chord progressions with very simple
heuristics. Chord progressions are given in MIDI and audio (mp3)
format:
Chord progressions sampled from the HMM model
Chord
progressions have been sampled from the HMM model with 36 hidden
states. The parameters have been learned on the same database with the
EM algorithm. One can hear that the excerpts are not ending on the
tonic chord most of the time. Also, chord sequences are likely from one
chord to another, but don't have a coherent global structure. Chord
progressions are given in MIDI and audio (mp3) format: