The Audio-based Covid-19 screening tool

Recall a time in India when 1,225 new COVID-19 cases were reported in 24 hours on March 31, 2022.

As per a report, only 28 cases are reported in the 24 hours, indicating a significant decrease in COVID-19 fatalities. The COVID-19 limitations were lifted, but it was still necessary to wear masks and keep social distance in public areas.

According to a study by the UK Alan Turing Institute, The Artificial intelligence sound classifier model was trained on different cough audio recordings to determine whether someone has Covid-19.

The reason for building an audio-based AI-powered Covid-19 app was to develop a quick and affordable way to test coronavirus. The UK Department of Health and Social was allotted two contracts of one million pounds to build the “Cough in A Box” for the government in 2021.

It could not correctly determine whether someone was infected with Covid-19 by listening to the sound of their coughs. The researchers state that AI can distinguish between coughs of Covid-19 and without it with 98.5% accuracy.

They also conclude that the model detects accurately based on user data like gender, age, etc. The researchers studied data from 67,000 people in which participants sent the recording of their cough, talking, and breathing along with Covid-19 nose and throat swab test results.

However, when researchers began to analyze the data, they found that accuracy was achieved by confounding (defined as a statistical phenomenon where the model trains with additional factors to correlate with the actual signals rather than the true signal itself.)

They conclude that the model performance is poor in detecting Covid by listening to cough.