Machine Learning in Automatic Music Chords Generation

Melodies and chords compose a basic music piece. Assigning chords for a music melody is an important step in music composition. This project is to apply machine learning techniques to assign chords to follow several measures of melody and generate pleasant music pieces. In real music chord assignment process, to choose which chord to use for a measure, musicians normally consider the notes in this measure and how chords are progressed. So our aim is to learn the relationship between notes and chords, as well as the relationship between adjacent chords, and use that to assign chords for a new melody. The input to our algorithm is a music piece with several measures. We then try different models to output a predicted chord for each measure. We will use basic models taught in the class (Logistic Regression, Naive Bayes, Support Vector Machine) as well as some advanced models (Random Forest, Boosting, Hidden Markov Model), and compare their performance for our problem.