AI Music Generation

This project aims to fuse 2 tracks of music together, using two methods. A Variational Auto Encoder and a Generative Adversarial Network. Included in this project are also:

• A website which showcases some examples as well as serves as an interface for both systems.
• A site scraper which was used to efficiently download midi samples from websites which offer them.
• A transposer which chages all midi file samples to the same key.

Description

Three approaches were taken to achieve this task:

1. Interpolation on Existing Music Passing 2 MIDI tracks through a Variational AutoEncoder.
2. GAN given all of the Music as a Training Set A MIDI Track produced by a Generative Adversarial Network trained on a particular Training Set.
3. New Music produced by GAN then interpolated by VAE Passing the 2 MIDI tracks produced by the Generative Adversarial Network through the Variational AutoEncoder.

Getting Started

Dependencies

• Python 3.8.8

• Flask version 1.1.2

• note_seq version 0.0.3

• magenta version 2.1.3

• Keras version 2.4.3

• Jinja2 version 2.11.3

• music21 version 6.7.1

• matplotlib version 3.4.1

• numpy version 1.19.5

Installing

To install all above libraries at one go

pip install -r requirements.txt

Executing program

• Open a terminal
• cd to current directory
• Run the following command

python -m flask run

• Open link given

Authors

Contributors names and contact info

• Paolo Bezzina - @PaoloBezzina
• Logan Formosa - @lowgunnn
• Yran Riahi - @YranRiahi
• David Dimech - @daviddimec
• Daniel Grech - @dangee303

Acknowledgments

• Music-VAE
• Magenta
• OpenAi Jukebox
• Rave Dj