Developed within the framework of the Computational Music Creativity Course (Music Technology Group, Barcelona).
The EnvironmentalDrumMachine.py Python file contains the source code of the framwework. The code is commented,
so the user can know what is done at each step. The main class and all the functions have a formatted docstring
where you can find the information of what is the main use of the functions, what are the expected parameters and
outputs.
The experiment.ipynb is basically hosting the interaction between the user and the application. Along the notebook,
you can find useful explanations of the project and how to play with it. Some examples are already provided.
The classification.ipynb provides an example code walkthrough to automatically build your environmental drum kit with
a Machine Learning model. The training and evaluation dataset is the
Freesound One-Shot Percussive Sounds Dataset.
The folder ./kit contains the Environmental Sounds Drum Kit generated using the Drum Part Classification Model.
This modeel was generated using sklearn, and the features were extracted using Essentia's Music Extractor. Please
note that the samples are automatically segmented using an energy threshold, aiming at removing the silent or
unvoiced regions from them. You can review the implementation of this approach in the
The ./patterns folder contains example patterns written in JSON format. Further detail and instruction to create
your own patterns is given in the notebook.
This project is not installable yet. It is also not robust to errors. In other words, it assumes that the drum patterns are written correctly, and that the folder structures and files are organized in the right way. While we increasr the robustness of the application, please make sure you write the patterns well and organize the sounds as they are organized now.
If you have doubts or problems reproducing the experiment, please contact me at genis.plaja01@estudiant.upf.edu.
Thanks:)