fredRan24 / OCSED_Utility

A tool which allows users to easily design, train and evaluate one-class Sound Event Detection machine learning problems

To use OCSED_util you must first create a virtual environment with the appropriate packages installes.

1. Follow the steps here to create an environment and load the requirements.txt: https://packaging.python.org/en/latest/guides/installing-using-pip-and-virtual-environments/

OCSED_util is a simple python class which you can implement to:

1. Preprocess data into binary matrix format for one-class SED purposes.
2. Train models
3. Evaluate models using the DESED evaluation set.

The DESED dataset is not included here, so you need to download it from here:

1. https://project.inria.fr/desed/
2. https://zenodo.org/record/3550599#.ZECNxnbMKHs

You should try to achieve the follwing file structure to work with OCSED_util as is:

. └── BARKSED/ ├── DESED/ │ └── dataset/ │ ├── audio/ │ │ ├── eval/ │ │ │ └── public/ │ │ │ ├── 1.wav │ │ │ ├── 2.wav │ │ │ └── ... │ │ ├── train/ │ │ │ └── synthetic21_train/ │ │ │ └── soundscapes/ │ │ │ ├── 1.wav │ │ │ ├── 2.wav │ │ │ └── ... │ │ └── validation/ │ │ └── synthetic21_validation/ │ │ └── soundscapes/ │ │ ├── 1.wav │ │ ├── 2.wav │ │ └── ... │ └── metadata/ │ ├── eval/ │ │ └── public.tsv │ ├── train/ │ │ └── synthetic21_train/ │ │ └── soundscapes.tsv │ └── validation/ │ └── synthetic21_validation/ │ └── soundscapes.tsv ...

Please see example.ipynb for examples of how to load, save, preprocess, train and evaluate your model/data