The Sound of Your Cough Could Help Screen For COVID-19

Will temperature checks and painful nasal swabs soon be a thing of the past? By analyzing the sonic features of a forced cough, a piece of cutting-edge artificial intelligence software shows promise at identifying people who have COVID-19, according to the results of a study conducted by a team of three researchers at the Massachusetts Institute of Technology (MIT). The September study was published in the IEEE Open Journal of Engineering in Medicine and Biology. 

Like all respiratory diseases, COVID-19 targets and attacks organs and tissues such as the lungs, the larynx (or voice box), and the trachea (or windpipe), limiting oxygen intake and causing corresponding changes in breathing and speech. In severe cases, these changes “may result in breathing difficulties that might take months to improve,” Katherine Herz, MPH, adjunct instructor of global health studies at the University of Iowa who was not involved with the study, tells Verywell, citing Johns Hopkins University. In mild cases, these changes can be too subtle for the human ear—but not for sophisticated (and super-sensitive) AI technology—to detect. 

"The sounds of talking and coughing are both influenced by the vocal cords and surrounding organs," Brian Subirana, PhD, director of the MIT Auto-ID lab and one of the authors of the study, told ScienceAlert. "This means that when you talk, part of your talking is like coughing and vice versa. It also means that things we easily derive from fluent speech, AI can pick up simply from coughs, including things like the person's gender, mother tongue, or even emotional state.”

How Was the Model Developed? 

The researchers developed the MIT Open Voice Brain Model (MOVBM), an AI-based “speech processing framework” that serves as a COVID-19 diagnostic test. The MOVBM relies on a set of five biomarkers, or traits commonly associated with a particular disease or disorder, to discern respiratory impairments of the characteristics of infection with COVID-19. These biomarkers include:

• Muscular degradation
• Changes in vocal cords
• Changes in sentiment/mood
• Changes in the lungs and respiratory tract

“The physical structure of the lungs and respiratory tract get altered with respiratory infections, and in the early days of the COVID-19 [pandemic], epidemiologists listened to the lungs while patients forced coughs as part of their diagnostic methods,” the authors wrote, describing the ways COVID-19 affects the quality of vocalizations. 

By creating a trilingual (English, Spanish, and Catalan) audio recording engine, the authors were able to collect audio recordings of forced coughs by 5,320 participants as well as any relevant medical information. Data from 4,256 of the participants was subsequently fed into the model to “train” it to distinguish between the forced coughs of those who had tested negative for COVID-19 and the forced coughs of those who had tested positive for COVID-19; data from the remaining 1,064 participants was used to test its ability to do so.

Overall, the model correctly identified 100% of asymptomatic COVID-19-positive audio recordings, 98.5% of all COVID-19-positive audio recordings, and 88% of all audio recordings. 

Does the Model Have a Future in Health Care? 

These results, the authors wrote, suggest that the MOVBM “has great potential to work in parallel with healthcare systems to augment current approaches to manage the spread of the pandemic.” They point to the model’s advantages over existing COVID-19-detection tools as evidence for their claim. Unlike current virology and serology tests, which cost an average of $23 each and take several days to process, for example, the MOVBM is totally free, provides instant results, and has a higher degree of accuracy to boot. 

Physicians see both functional and practical obstacles to its widespread implementation, however. Joshua O. Benditt, MD, professor in the division of pulmonary, critical care, and sleep medicine at the University of Washington School of Medicine, tells Verywell that “it is an interesting idea but would have to be tested on a population of people who are symptomatic but with another disease.” 

“In my mind, the real question is, ‘Can this program differentiate the cough of someone with COVID-19 from someone with the common cold (also [a] coronavirus), influenza, bacterial pneumonia, and other commonly seen conditions?’” he says. 

Herz believes that the model has potential but that its technological sophistication may prove its downfall. 

“While the cough recordings sound hopeful, it is not clear how much time will be needed to get approval from the FDA, to produce more machines capable of analyzing cough patterns as the study describes, as well as training people so [they] are able to use the equipment properly so that there are as few false positives and false negatives when people are tested,” she says. After all, it is much simpler to swab the inside of someone's mouth than it is to run an analysis on an audio recording.

The information in this article is current as of the date listed, which means newer information may be available when you read this. For the most recent updates on COVID-19, visit our coronavirus news page.