Researchers are working on machine learning systems to identify COVID-19 cases by the sounds of a person's cough.

One system has demonstrated a high success rate in detecting COVID-19 in people with no physical signs of the disease. Such a tool could be important in the fight against COVID-19, which can be spread by people who do not even know they are infected.

Researchers at the Massachusetts Institute of Technology, MIT, recently published a paper reporting results of the system.

The team created an artificial intelligence (AI) model to examine the sounds of people who produced a forced cough. The sounds were collected from people who recorded them on computers or mobile devices. The individuals were also asked to provide information about any symptoms they were experiencing, as well as whether they had been officially tested for COVID-19.

People then sent the recordings and data to researchers through the internet or their devices. Researchers reported they had received more than 70,000 recordings, amounting to about 200,000 individual cough examples. The team then trained the model on the cough sounds, as well as spoken words.

When the new cough recordings were fed into the system, it correctly identified 98.5 percent of coughs from people confirmed to have COVID-19, the researchers reported. The model also detected 100 percent of coughs in people who reported they had tested positive for the virus, but had no signs of the disease.

One of the project's leaders is Brian Subirana, a research scientist in MIT's Auto-ID Laboratory. Subirana and his team had already been developing AI models to examine forced-cough recordings to search for signs of Alzheimer's disease. Such signs can include changes in personality and memory loss, but Alzheimer's can also cause nerve and muscle problems, including weakened speech.

The MIT team says its latest model trained to identify Alzheimer's disease from cough sounds had also shown good progress as a possible way to help detect the condition.

So when the coronavirus pandemic developed, Subirana told MIT News, he thought the same model structure might work for COVID-19. This is because there was evidence that COVID-19 infected individuals may also experience voice muscle weakness.

Subirana said the researchers discovered "a striking similarity" in the ability of the model to detect Alzheimer's and COVID-19. The experiment showed that the way a person produces sound changes if they are infected with COVID-19 even if no physical signs are present, he added.

The team says it is working to develop "a user-friendly app" that could be used on a wide basis to detect COVID-19 cases. This would make it possible for users to cough into their phone and receive immediate information on whether they might be infected and should seek an official test.

The effective use of such a tool could also "diminish the spread of the pandemic if everyone uses it before going to a classroom, a factory, or a restaurant," Subirana explained.

U.S. researchers at Pennsylvania's Carnegie Mellon University are also using machine learning methods to develop a "voice-based testing system for COVID-19." That system also uses recordings of coughs – as well as some vowel sounds and the alphabet – to identify "signatures" of the virus, the Pittsburgh Post-Gazette reported.

And in Britain, a similar project is being carried out by engineers at the University of Cambridge. Researchers working on that system reported in July they had created a machine learning tool that could correctly identify COVID-19 cases based on cough and breath sounds. Those models performed with a success rate of about 80 percent in laboratory tests, the team reported.

I'm Bryan Lynn.

Bryan Lynn wrote this story for VOA Learning English based on reports from MIT, the University of Cambridge, Carnegie Mellon University and the Pittsburgh Post-Gazette. Susan Shand was the editor.