Cough-analyzing AI detects asymptomatic COVID-19 infections

It can hear sounds in a forced-cough that the human ear misses.

Asymptomatic COVID-19 infections have been a huge barrier to getting the pandemic under control.

Experts estimate that between 20% and 45% of COVID-19 cases never experience symptoms — and almost everyone has a few days when they are contagious before symptoms appear. Because people don’t feel sick, they don’t get tested or quarantine — but they do spread the disease to others.

Now, researchers at MIT have developed an AI that can detect asymptomatic COVID-19 infections just from the sound of a person’s forced-cough — a development that could help stop these “silent transmissions.”

Detecting the Coronavirus

To the human ear, forced-coughs from people with asymptomatic COVID-19 infections sound the same as those from healthy people, but there are subtle differences that AI can detect.

To train it to spot these differences, MIT researchers started by setting up a website in April where people could submit audio recordings of themselves forcing a cough, as well as information on their COVID-19 status and symptoms (if any).

They received more than 70,000 recordings, including around 2,500 from people with COVID-19, some of whom were asymptomatic. They built a dataset containing all of the coronavirus-positive recordings and about 2,500 recordings from healthy people.

After using about 4,000 of the recordings to train their AI, they used the remaining clips to test its ability to distinguish between healthy people and those with COVID-19.

The AI was able to correctly identify 98.5% of the coughs from people with COVID-19 and 100% of those from people with asymptomatic COVID-19 infections, according to the researchers’ study, published in the IEEE Journal of Engineering in Medicine and Biology.

“We think this shows that the way you produce sound changes when you have (COVID-19), even if you’re asymptomatic,” co-author Brian Subirana told MIT News.

Catching Asymptomatic COVID-19 Infections

The MIT researchers didn’t have to start from scratch when developing their cough-analyzing AI for COVID-19.

The project began before the pandemic as an effort to detect Alzheimer’s by listening for biomarkers of the disease, including changes in lung performance and vocal cord strength, in a person’s cough.

This tool could diminish the spread of the pandemic if everyone uses it.


Brian Subirana

Now that they’ve shown that the AI is capable of detecting the coronavirus, too, they’re working to develop an app that could serve as a screening tool for asymptomatic COVID-19 infections.

People could log into the app, force a cough into their smartphone’s microphone, and find out if they’re likely to have an infection. If so, they could then visit a clinic for a formal COVID-19 test.

“The effective implementation of this group diagnostic tool could diminish the spread of the pandemic if everyone uses it before going to a classroom, a factory, or a restaurant,” Subirana said.

We’d love to hear from you! If you have a comment about this article or if you have a tip for a future Freethink story, please email us at tips@freethink.com.

Up Next
deepfakes in sassy justice
Subscribe to Freethink for more great stories