A Machine Learning Approach to Support Music Creation by Musically Untrained People
1. A Machine Learning Approach to Support
Music Creation by Musically Untrained People
Tetsuro Kitahara and Yuichi Tsuchiya
Nihon University, Tokyo, Japan
How can we use computing technology to
encourage music creation by novices?
Issue 1
Human interface
Users need to be able to
input their musical ideas
in an intuitive, easy way
Issue 2
Automatic music generation
System needs to be able to generate musical
pieces from abstract / incomplete inputs
Powered by machine learning
Our proposal: Melody editing based on melodic outline
â Melody is represented as a curve
â User edits the melody by redrawing the curve
â Usable when the user is not satisfied with
melodies generated by a music composer
Idea 1 Melodic outline
â Estimate a sequence of notes satisfying:
â Closeness to the outline
â Musical appropriateness
Idea 2 Hidden Markov model
Mutual transformation between melody and melodic outline
Hidden Markov model for melody generation
Transform melody to melodic outline
Pitch trajectory
Fourier transform
Inverse Fourier
transform Save for later use
Extract low-
order coeffs.
High-order coeffs. of
original melody
Low-order coeffs. of
Fourier transform
Inverse Fourier transform
Do Mi Fa So Ti Ra So Do Re Mi
Hidden Markov model
Transform melodic outline to melody
Edited melodic outline
Key idea
Seq. of note numbers (musically appropriate)
Emit (with random deviation)
Seq. of continuous pitches in melodic outline
48 49 84,,,
Each state represents
each note number
Observed signal
(1-dim continuous value)
Hidden states
Emission prob.
For state si
,
i Note
number
pitchi
Closeness to userâs
melodic outline
Transition prob.
Note name unigram
Interval unigram
C D E F G A B
0
0.2
0.4
P1 M2 M3P4 P5 M6 M7
0
0.2
0.4
Musical appropriateness