Details of the full project can be found in Covid Patient Monitoring.pdf file

For Covid patient monitoring, measuring cough rate is a cruical and important task. By monitoring cough rate we can decide different conditon of the paitient. In this project we tried to measure cough rate in real time.

For cough rate measuring we trained a deep learning model which can effectively detect cough from the audio input. Details of the dataset can be found here . We tried to build a robust model sothat it can differentiate cough from other sounds like sneeze, talking, background noise, intrumental noise etc .
Detailed explanation of the dataset preparation can be found here.

We used YAMNet as our base network which is a modified version of Mobilenet_v1 for audio data classification. As our works is a work for audio classification, whether the incoming audio is cough or not, we used some layer on top of the base YAMNet, as YAMNet is designed for classifying 521 audio event classes.

We used AudioSet dataset released by Google as our training data. Main Audioset consists of 521 classes which is firmly annotated. We only used a fraction of that dataset. We used cough data as positive event and sneeze, sniffle, breathe, hiccup, gasp, silence and speech audio data as negative event.

Dataset Distribution:
Train:-
Cough= 4926
Other= 5530
Validation:-
Cough= 1231
Other= 1777

We trained our model got around 88.43% accuracy on our validation dataset.
There are some results of our model.

For running inference run final_cough_rate_notebook.ipynb notebook.

N.B: pyfirmata module was used to connect the device with arduino to show the results in a LCD display also blink an LED for extreme case.