Music Analysis and Generation Supervisors: Jon McCormack & Lloyd Allison Final Presentation Oliver Ng (B.DigSys)
Overview Music representation: Music representation: –Notation –Waveform Applications Applications –Games –Interactive environments
Aims of research Analysis: Finding a set of rules that uniquely identify a genre of music Analysis: Finding a set of rules that uniquely identify a genre of music Generation: Creating similar music in the same genre based on rules Generation: Creating similar music in the same genre based on rules Deliverables: Sample music Deliverables: Sample music
Basics Note Note –Pitch –Duration Key signature Key signature Time signature Time signature Tempo Tempo
Basics “12 tone theory” – 12 semitones/octave “12 tone theory” – 12 semitones/octave
Basics Key signature: Key signature: Absolute vs Relative to a tonic key Absolute vs Relative to a tonic key
Basics Key signature: Key signature: Absolute vs Relative to a tonic key Absolute vs Relative to a tonic key
Previous Research Analysis Analysis –Template key matching Generation Generation –Stochastic processes –Order N Markov Models
Methodology - Use of 2 separate entities or “agents”
Implementation MIDI MIDI –Events (Tempo, Note On, Note Off) –Tracks (Monophony, Polyphony) Haskore (MIDI routines) Haskore (MIDI routines) C++, Unix shell scripting C++, Unix shell scripting
Analysis Template key matching Template key matching –Base 12 –Use of 24 preset chords (12 major, 12 minor)
Analysis Template key matching Template key matching –Base 12 –Use of 24 preset chords (12 major, 12 minor)
Analysis Template key matching Template key matching –Base 12 –Use of 24 preset chords (12 major, 12 minor)
Analysis Template key matching Template key matching –Base 12 –Use of 24 preset chords (12 major, 12 minor)
Analysis (continued) Relative vs absolute key Relative vs absolute key 2 tiered model 2 tiered model –Bottom up Note level Note level Bar level Bar level –Ideal vs actual chord definitions –1 st order Markov Model constructed on both tiers. Monophony and polyphony Monophony and polyphony
Generation Top down Top down –Selection of chords per bar –Selection of notes based on chosen chords
Results Fully handwritten Fully handwritten Partially handwritten, partially generated Partially handwritten, partially generated Fully generated Fully generated
Conclusions and further work Sounds fairly similar at a bar level Sounds fairly similar at a bar level –(2 nd tier) Add 3 rd tier Add 3 rd tier Extend order of Markov Model Extend order of Markov Model –(training data) Feed in multiple pieces Feed in multiple pieces
Further information Project website: Project website: “ Some artists claim their best creative work is done under the influence of a large variety of chemical substances ’’ - McCormack