COVID-19-Detection-using-NLP

Presentation Link: https://www.youtube.com/watch?v=uSDq2PNSypE

In this blog, we have focused in distinguish the cough to determine if the person is infected with coronavirus or not.

We start by importing the packages and configuring some settings.

Loading Data

We read the data set coughvid from the virufy repo.

It contains a CSV file that contains the patient ids and some information about the patient health.

Visualizing dataset

Audio Preprocessing

Using the trim method from libsora library, we have implemented the trim_silence function to trim silence from the audio signal and its intervals.

Get mel spectrogram using libsora method and convert it to DB and save the result as image .png.

Get raw MFCC feature and get the label of every patient.

Then implement and filter data set to extract necessary features in JSON files.

The JSON file contains a dictionary of dictionary patient ID, mel spectrum image. MFCC coefficients, and labels.

Approach 1 using MFCC Coff and labels:

Work on JSON files and create a CSV file contains only patient ID, rawMFCC, and labels.

- Model Architecture:

Implement logistic regression architecture:

And the result is :

Approach 2:

In this approach, we have worked on images generated from mel scale and labels

-Preprocessing to extract images and their labels:

Using ImageGenerator to overcome unbalanced data:

Then, Implement a CNN model

Using Mobilenet architecture and adam optimization

The result is :

Then using Resnet 50 and use 20 epochs to train :

Getting the result :