This project is aimed at creating a deep learning model using Long Short-Term Memory (LSTM) for generating melodies. The model is trained on a dataset, specifically from the Alsace (eastern France) of MIDI files to learn the patterns and relationships in musical sequences.
The dataset used on this project is from Alsace, a historical region in France where they have plenty of collected compositions in .krn files that can be accessed using any music notation softwares.
The following packages required to access and manipulate musical notations:
Start by cloning the repository to your local machine
git clone https://github.com/<username>/lstm-melody-generator.git
Set the MuseScore3 path to view the notations of .krn file sets
# set the default musicxmlPath file reader to MuseScore3.exe
us = m21.environment.UserSettings()
us["musicxmlPath"] = 'C:/Program Files/MuseScore 3/bin/MuseScore3.exe'
Towards the preprocessing, provided are .txt files from the .krn file sets and are then tokenized stored within mapping.json as file_dataset being the final output of preprocessing
python preprocess.py
For training, an .h5 file is produced for creating the melody
python train.py
To generate the model, run the following command; the seeds can be configured to the formatting required
python melody_generator.py