COVID-19 Identification ResNet (CIdeR)

Repo for the classification of COVID-19 using cough and breath audio samples from End-2-End COVID-19 Detection from Breath & Cough Audio.

Schematic of the CIdeR architecture

Contents:

main --> Script for orchestrating training and testing of model.
data_preprocessing --> Loads and and performs all preprocessing of data.
conv_model --> code for CIdeR Train/Validaiton/Test splits --> The train/validation/test splits used in the paper.
dev_test_train_check --> Tests to make sure no leakage of users between train/val/test splits occurs.
args --> The hyperparams of the final model.
create_csv --> code used to create train/validation/test splits. note: running this file will overwrite the splits used in this work and so if any comparision is to be made, do not run this script

Replication of Results

For privacy reasons we do not release the trained models however, we can provide them on request. Requests to harry.coppock@imperial.ac.uk

To train or evalute the model first install the required packages:

$ pip install -r requirements.txt

To train and evaluate the model with the hyperparams defined in args.txt simply run:

python main.py

If you would like to evaluate an existing trained model:

python main.py --do_train False --saved_model_dir pathtothesavedmodel 

Contributing

Please feel free to contribute to this repo:

1. Fork this repo
2. Create a branch
3. Commit and push your changes
4. Create pull request
5. Someone will review your changes and merge them.

Questions & Issues

Create an issue and someone will respond!

Cite

Please cite our paper

@article{coppock2021end2end,
      title={End-2-End COVID-19 Detection from Breath & Cough Audio},
      author={Harry Coppock and Alexander Gaskell and Panagiotis Tzirakis and Alice Baird and Lyn Jones and Björn W. Schuller},
      year={2021},
      eprint={2102.08359},
      archivePrefix={arXiv},
      primaryClass={cs.SD}
}

If you find this repo useful, please consider giving it a ⭐