robsdedude/melody-composer

Using higher Order Markov Modells and Clustering to compose melodies similar to a training data set

This program is part of my bachelor thesis which can be found on arXiv. It's a machine learning algorithm that generates monophonic melodies similar to those of a phrase annotated data-set. In- and output are midi files (i.e. machine readable music).

Setup

Use virtualenv

virtualenv venv
source venv/bin/activate
export LC_ALL=C
pip install -r requirements.txt

Troubleshooting

On Ubuntu based systems it can happen, that there are header files missing. Do: sudo apt-get install liblapack-dev libblas-dev It can also happen, that you need to install a fortran compiler. Do: sudo install gfortran After fixing these things re-run pip install -r requirements.txt each time.

Trainings Data

Download the MTC-FS data set and unzip it into the folder MTC-FS-1.0

Then run the preporcessing

mkdir -p MTC-FS-1.0/good_midi  # create folder to put pre-processed midi files in
source venv/bin/activate       # activate virtualenv
python preprocess.py           # run preprocessing
deactivate                     # deactivate virtualenv

Run the program

Start the virtual environment with source venv/bin/activate To general start the program go to this directory and run python main.py More information: python2 main.py --help

Other things

To see how the clustering is working have a look at main() in contour.py