Project repository for EEE4114F 2024, DGMROB001 and PLLTHI032. University of Cape Town (UCT). Instrument classification using convolutional neural networks on audio spectrograms.
Honestly, I don't really recommend navigating this repository. As is probably evident, by the end of the project, the folder structure was a bit of a mess. That being said - if you really want to:
Code: contains various pieces of code for generating spectrograms and other forms of feature extraction, as well as implementing the machine learning algorithms. The most interesting notebooks are probably spectrograms.ipynb (spectrogram generation) and learning.ipynb (machine learning algorithms).
Training-data and Test-data: Arguably the most interesting folders to look at - contain the sounds, spectrograms, and MFCCs used in the process of developing the model. Sounds were largely taken from freesound.org and pixabay.com. Piano sounds were recorded from Helm tytel.org/helm/, an open-source digital synthesizer.
Report: LaTeX source code for the report. Final pdf is viewable in the main repository of the project.
Future UCT students who may come across this repository, absolutely do not have permission to plagiarise from our report. We hope this project may provide inspiration for future students or other curious individuals, but be cautious about directly copying any part of the project and claiming it as your own work.
This repository was made for a specific course project, and will be retired after the project's conclusion. Pull requests will not be accepted.
Goodbye, and thank you for reading this readme. Have a lovely day :)