🔊 NEXUS Cough Classification

This repository contains a training pipeline for a Cough classification model using Various Model. It supports audio augmentation via the RIRS noise dataset for realistic room simulation.

---

🏆 Getting Started

Model	Accuracy	Balanced Accuracy	AUC	ROC AUC	F1 Score Accuracy
LSTM	0.87	0.88	0.95	0.88	0.87
ResNet101	0.91	0.91	0.97	0.91	0.91
InceptionV3	0.92	0.92	0.98	0.92	0.92

---

🚀 Getting Started

1. Clone the Repository

git clone https://github.com/arkiven4/NXSCough
cd NXSCough

2. Preparation

Install dependencies with:

pip install -r requirements.txt

Before training, ensure that the path to your dataset is correctly set in your config in the config/lstm.json folder:

"db_path": "/run/media/fourier/Data1/Pras/Database_ThesisNew/data/"

This folder must contain:

• somedatasets.csv: includes audiopath, labels
• audiodata folder: contains the actual audio files or processed data

/run/media/fourier/Data1/Pras/Database_ThesisNew/
├── somedatasets.csv
└── audiodata/
    ├── file1.wav
    └── file2.wav

Update the path according to your environment before running the training script.

3. Train the Model

python ztrain_nonssl.py

Training logs will be saved to the ./logs/ directory. You can Check with this command:

tensorboard --logdir ./logs/lstm_try1

Open the provided localhost link in your browser to monitor loss, accuracy, and other metrics during training.

After training is complete, please check the logs folder.

🚀 Getting Started For Singularity User

Follow the steps below to clone the repository, build the Singularity image, bind necessary folders, and execute your application.

1. Clone the Repository

git clone https://github.com/arkiven4/NXSCough
cd NXSCough

2. Build the Singularity Image

singularity build nxscough.sif nxscough.def

3. Configure Config

Before training, ensure that the path to your dataset is correctly set in your config like this:

"db_path": "/mnt/data/"

4. Run Training

sudo singularity exec --bind /run/media/fourier/Data1/Pras/Database_ThesisNew/:/mnt/data --nv nxscough.sif python3 ztrain_nonssl.py

The path /run/media/fourier/Data1/Pras/Database_ThesisNew/ replace with path on your host machine and must contain:

/run/media/fourier/Data1/Pras/Database_ThesisNew/
├── metadata_combine.csv
└── CombineData/
├── file1.wav
├── file2.wav
└── ...

• metadata_combine.csv: A CSV file containing metadata about the audio files.
• CombineData/: A folder containing .wav files used for training.

These files will be mounted to /mnt/data inside the container.

Optional: Data Augmentation with RIRS

To apply room impulse response (RIR) augmentation:

Download the RIRS dataset from: 👉 https://www.openslr.org/28/

Extract the contents and place them in a suitable path used by your config or augmentation script.