Cough Detector

I developed a web-based cough detection app which is available at https://cough-detector.onrender.com/.

The code for the web app can be found at https://github.com/AVS1508/cough_detector.

NOTE: I have deployed the app on Render under their free plan so if the web dyno spins down after inactivity, it may take up to 5 minutes for the app to spin up due to low CPU resources provisioned.

Due to the unreliable nature of the spin-up/spin-down mechanism, I have recorded a demonstration video of the app available at https://youtu.be/JdT60zeE36I

Implementation Details

The overall app is built using Flask with the main app file being app.py which is served on a remote server using the server.py file.
The learned cough detection models are stored in a .npz structure and the best model (regular model with $\lambda = 6$) is fetched from that structure in predict.py. The file also contains the necessary code for sparsifying the input features and model weights with threshold $\theta^i < 0.01$, for extracting features using librosa.feature.melspectrogram and for making predictions using the prediction function.
The app records an audio sample from any available user media device using static/js/main.js and then sends it over to an endpoint for the Flask app which then extracts features, applies the model, and returns the prediction. This prediction is then qualitatively presented on the user interface for the user to detect the presence of coughs.