The goal of this project was to predict the notes given the beat and note times of classical music songs, and generate new songs using the predicted notes and the timings.
I used a decision tree regression model, and fit to the the musicnet dataset.
Make sure to press the unmute button to hear the audio!
demo2.-.2225.mov
Result for Prelude and Fugue in E-flat major, BWV 852 by Bach (Played with steel drums)
demo3.-.2303.mov
Result for Prelude and Fugue in D major, BWV 850 by Schubert (Played with music box)
demo1.-.1757.mov
Result for Piano Sonata in C minor, D. 958 by Schubert (Played with grand piano)
sklearnpandasnumpymidiutiljupyter-notebook
- Navigate to source code folder
- Create folder
input - Dowload
musicnet, extract to./input/ - Download
musicnet_metadata.csv, extract to./input/musicnet/
- Navigate to source code folder
- Run jupyter notebook
- Open ai_generated_midi.ipynb
- Edit song_id to a song id in the musicnet dataset of your choice
- Run all cells
- Open in the resulting
output.midiin a midi player (volume warning).