Deep Learning for Automatic Melody Harmonization

Beginner music composers often face difficulty in harmonizing a melodic idea. Particularly, if they lack music theoretical knowledge. It is well-understood that chord successions follow patterns. With some limitations, artificial intelligence algorithms can capture those patterns. A melody “harmonizer” model proposes harmonizations for the user’s melodies. The suggestions are often based on learned patterns from existing musical compositions, for example, the chorales by J. S. Bach. A model may learn patterns found in Bach’s compositions, applying them in new music. An alternative approach may be to condition the music generation with certain constraints. For example, performing a music-theoretical analysis of the melody first. The model generates harmonies based on the tonal analysis of the melody. This project consists of developing a melody harmonization system using the latter approach. The system is intended to operate with Sibelius and Pro Tools, two software applications developed by Avid Technology.

Faculty Supervisor:

Ichiro Fujinaga

Student:

Néstor Nápoles López

Partner:

Discipline:

Music

Sector:

Information and cultural industries

University:

McGill University

Program: