Analyzing Cough Sounds for Triaging Covid-19 Patients

As many AI and Machine Learning enthusiasts must be aware of the fact that many Researchers are trying to develop an app which can correctly help patients to detect whether they are Corona Positive or not.Though these apps not yet accurate enough to detect Corona Patients because researchers are yet collecting training data and asking Corona Patients to record their cough sounds and upload them on the website.As we are all aware that respiratory sounds are known to have diagnostic properties but there are also evidences that the Coughing Sound can also help in Covid detection.You all must be knowing about current Australian Treasurer, Josh Frydenberg who got indications about being Corona Positive when suddenly he started coughing in the middle of his speech in the Parliament.

So, I have tried to develop a model which can make use the Patient's Coughing Sounds and convert it in Mel-scaled Spectrogram format which if saved in an image format can be then fed to an image classifier to classify whether the patient is Corona Positive or not.

A Corona Patient cough sounds like this and A Normal Patient cough sounds like this.

A Melspectogram of the cough sound of Covid-19 Patient.

A Melspectogram of the cough sound of Non Covid-19 Patient.

Since,no such Coughing Sound dataset has been released.So,I collected these sounds on my own and the size of dataset is relatively very small and I am still working on my model to include more features and more data.But,my main intention was to develop one such model because there have been many efforts of detecting Covid Patients through their Chest X-Ray but it would not be of much help to the patients as they have to get out of their homes to get their X-rays done and then there are very strong chances that he/she will eventually end up propagating the virus.So,one such model is the very needed for Covid detection.

Many of you might be wondering that why I changed the audio data to image data and then worked on the image data.Yes,I could have worked with the audio data and made use of RNNs as well but I am more of a Computer Vision enthusiast so I took the Computer Vision approach but I will also try to develop one such model which can make use of audio data without needing it convert it to an image data.

The link to my model code is given below and you can have a look at it and please feel free to give any suggestions.

My Model Code If you are unable to open the link follow this: "https://cainvas.ai-tech.systems/notebooks/details/?url=cainvas-static.s3.ap-south-1.amazonaws.com/media/user_workspace/arya.ashish/Covid_Cough_Detection.ipynb"