This project is part of the Advanced Machine Learning course at ENSAE. Its goal is to train a model that can recognize what instruments are playing in a musical audio. It will then have some other refinements.
We used various data for tests, but the one we finally used for training is the IRMAS dataset: https://www.upf.edu/web/mtg/irmas We couldn't put it directly on github as it is too big. Once it has been downloaded and put in data/original_data, the scripts in /helpers/dataset_creation/dataset_specific can be used to put them to the right format. The mian.py script of /helpers/dataset_creation/general_features must also be run to generate the split.jsonl files that tell for each file if they are to be used for testing, training, validation or not at all.
All the code of this repository was made by us.