RANDrumizer leverages an LSTM-VAE neural network to randomly generate symbolic drum patterns, each corresponding to a single 4/4 measure. Although it is a relatively simple model with limited performance, it serves as a functional demonstration of rhythmic sequence generation, providing a basis for more sophisticated approaches.
The code in this repository is contained in a single Jupyter Notebook, namely main.ipynb, which builds, trains and tests the neural network model. To demonstrate this process, a custom dataset compiled from numerous MIDI drum tracks and provided in the Files directory is utilized.
This project was developed exclusively within the computing environment of Google Colaboratory, thus the specific version of Python used and any package requirements are subject to the up-to-dateness of the service at the time of development. To run the notebook in your own working environment, please refer to the associated code sections and make any necessary adjustments.
RANDrumizer © 2024 by Alexandros Iliadis is licensed under the MIT License.
A short and simple permissive license with conditions only requiring preservation of copyright and license notices. Licensed works, modifications, and larger works may be distributed under different terms and without source code.
See the LICENSE.md file for more details.