This is a music machine learning project by Andrew Colabella.
This uses Midicsv/Csvmidi and text generation with LSTM recurrent neural networks.
MidiCSV was created by John Walker https://www.fourmilab.ch/webtools/midicsv/
The LSTM Recurrent Neural Networks is based off of https://machinelearningmastery.com/text-generation-lstm-recurrent-neural-networks-python-keras/
How to use:
In the top level directory, there exists a shell file called "runall.sh". This is the master script that runs everything.
All midi files to be used as input should be placed in the inputfiles/midis directory
The runall.sh script will call midicsv on all midi input files, converting them to csv. These csvs will be placed in the directory inputfiles/csvs. These will then be converted to text files using the editCSV python script. The editCSV python script reduces the csv files to just the instrument tracks and replaces descriptors such as Note_on_c and Note_off_c with integers, thus creating a text file with only integers, brackets ('[', ']'), and commas. This allows for a minimal amount of characters the recurring neural network needs to learn while still maintaining to complex data that represents these musical tracks.