COVID-19 is an invisible threat, but is it silent as well? Researchers from MIT’s Lincoln Laboratory just released a new study that suggests it’s possible to identify asymptomatic carriers of the coronavirus via nothing more than their speech signals.
How is this possible? Milder infections like the flu or a common cold make people’s voices sound more nasally or congested all the time, and the same vocal distortions are often observed among symptomatic COVID-19 patients. Now, by closely analyzing and processing recorded speech signals, the team at MIT has detected some very subtle vocal changes among asymptomatic COVID-19 patients.
These changes, or “vocal biomarkers” are undetectable to the naked human ear without technological assistance. Nevertheless, the study’s authors believe these findings may lay the groundwork for a new way of detecting active COVID-19 infections among asymptomatic carriers. Such a screening process could be achieved using a mobile app.
So, what exactly are these super subtle voice changes? The fluctuations are believed to stem from disruptions COVID-19 causes in the muscle movements of carriers’ articulatory, laryngeal, and respiratory systems.
“I had this ‘aha’ moment while I was watching the news,” says Thomas Quatieri, a senior staff member in the laboratory’s Human Health and Performance Systems Group, in a news release.
Quatieri has been researching vocal biomarkers of disease for the past decade. In the past, his work has primarily focused on vocal signs indicating neurological problems such as Parkinson’s and ALS. Recently, however, he had an epiphany while watching the news one evening; there are tons of easily accessible audio clips of asymptomatic COVID-19 carriers.
This realization led him to wonder if vocal biomarkers may also exist for COVID-19. It’s fairly easy to conclude that there are vocal signs of COVID-19 among symptomatic patients, as respiratory inflammation affects one’s ability to talk (loudness, pitch, steadiness, etc). Beyond those obvious cases, though, Quatieri and his team wanted to see if they could find any subtler signs of the coronavirus in asymptomatic carriers’ speech signals.
To that end, researchers searched through various YouTube clips of celebrities who had given interviews while coronavirus positive but asymptomatic. They settled on five unnamed celebrities, and then gathered a bunch of their interviews, from both before and after contracting COVID-19.
A series of algorithms were then used to single out and compare various vocal components of the celebrities’ speech patterns over time.
“These vocal features serve as proxies for the underlying movements of the speech production systems,” comments Tanya Talkar, a Ph.D. candidate in the Speech and Hearing Bioscience and Technology program at Harvard University.
For the research, the five subjects’ vocal amplitude (loudness) was considered a sign of respiratory movement. Larynx movements (vocal cord stability) were measured via pitch and the steadiness of said pitch, and articulatory movements (tongue, lips, jaw, etc) were measured using “speech formants.”
Speech formants are frequencies measuring how one’s mouth shapes sound waves to form vowels and consonants, and vocal quality. The researchers’ general hypothesis was that inflammation caused by COVID-19 results in vocal muscles becoming “stuck together” in a sense, ultimately leading to less complex movements.
“Picture these speech subsystems as if they are the wrist and fingers of a skilled pianist; normally, the movements are independent and highly complex,” Quatieri explains. Building off this example, when COVID-19 enters the equation, it forces the pianists’ wrist and finger movements together, causing the musician to play a simpler melody.
Across all of the analyzed recordings, researchers looked for signs of this “coupling” over time. Sure enough, a decreased complexity of vocal movements was observed during interviews recorded with asymptomatic celebrities after they had tested positive for COVID-19.
“The coupling was less prominent between larynx and articulator motion, but we’re seeing a reduction in complexity between respiratory and larynx motion,” Talkar says.
All in all, these initial findings strongly suggest that asymptomatic COVID-19 carriers experience slight changes in their vocal patterns that can be identified using the right approach. At this point, however, the study’s authors say more research is needed before this can be definitively agreed upon. Moving forward, they’ve already started collaborating with some colleagues from Carnegie Mellon University on analyzing more voice recordings from COVID-19 carriers.
Researchers are also looking into possibly incorporating this approach into a few different mobile applications.
“A sensing system integrated into a mobile app could pick up on infections early, before people feel sick or, especially, for these subsets of people who don’t ever feel sick or show symptoms,” concludes research group leader Jeffrey Palmer.
The full study can be found here, published in IEEE Open Journal of Engineering in Medicine and Biology.