Is a Voice Test for COVID Coming to a Smartphone Near You?
The race to develop an app that provides vocal screening for COVID-19.
Posted Dec 06, 2020 | Reviewed by Devon Frye
For anyone that has experienced a cotton swab up the nose for a COVID test, the idea of a non-invasive way to help determine whether you should visit your grandmother over the holidays probably sounds pretty good. And, if the companies that have been working on vocal biomarkers of COVID have their way, soon that may be a reality.
This past spring, several separate research reports from universities and the private sector suggested that computer-based voice analysis that looked at the sound waves created when speaking or coughing could help identify people with and without COVID. Promising early results suggested that the sound of your voice indeed held significant clues to your health status.
While these studies merely pointed to its potential as a diagnostic tool, teams working with the technology are now getting closer to making it available to the public. Soon, repeating a sentence or counting from 50 to 70 into a smartphone app may be an easy way to detect changes in the respiratory system associated with COVID infection. In other words, it could provide an early warning system to better determine if you should skip going to dinner at the neighbors or sending your kid to school.
Turning Research Into Reality
Several different research and industry teams are racing to develop the artificial intelligence (AI) as a usable tool, namely a U.S. and Israeli partnership, Vocalis Health, as well as researchers from Cambridge University and, independently, from Carnegie Mellon and MIT. By collecting voice samples from thousands of voice volunteers that were COVID negative and COVID positive, the scientists have been utilizing machine learning processes to identify a voiceprint for COVID infection.
The result? A screening tool in the shape of a smartphone app that just requires a few voice samples to provide an assessment of one's risk of being infected within a minute. This technology is, of course, not a replacement for a doctor’s visit or the ever-so-enjoyable nasal swab but may help identify those who should go for further testing.
So far, research suggests the algorithms from the varied projects are successful in detecting COVID-19 positive patients. When used in concert with a symptom questionnaire, the Cambridge, MIT, and Vocalis teams all report accuracy rates of 80 percent or higher in correctly assessing COVID status. Even more promising, the technology has been able to successfully detect even asymptomatic carriers from coughs or voice samples. Though not yet available to individuals, the technology is now being further tested in larger pilots and also by employers to help screen employees to encourage a safe return to work.
But before they can make the app available for general use, Cambridge, Vocalis Health, and Carnegie Mellon need to gather more voices to help fine-tune the technology. People can help speed up bringing this technology to practical use by volunteering their voice, as the projects rely on crowdsourcing voice samples to provide a large enough training set.
The Link Between Disease and Voice
This is not the first attempt to use AI to identify disease from a voice. Previously, scientists found that vocal markers were an extremely effective way of diagnosing and following patients with Parkinson’s disease. Other research suggests great promise in detecting coronary disease as well as mood and bipolar disorders.
Vocalis Health, one of the companies working on the COVID biomarkers, first applied voice analysis to identify shortness of breath in COPD patients, then pivoted to put similar AI to work to develop a working vocal biomarker for COVID-19. They also just announced a partnership with the Mayo Clinic to develop voice biomarkers for pulmonary hypertension.
This emerging field of vocal diagnostics offers a novel, cheap, and easily accessible approach that could have far-reaching application for a wide variety of diseases far beyond COVID. In fact, subtle voice changes have already been identified for conditions ranging from Alzheimer’s, autism, and obstructive pulmonary disease to Parkinson’s and depression.
For now, though, having a little help in the form of a quick screening tool for COVID that would allow a bit more confidence as we get back to work and visit with family is more than enough. And, as for the future of voice detection, it used to be listening to the sound of someone’s voice was just about paying attention. But one day soon, it might make all the difference for disease diagnosis and intervention, and with nary a cotton swab or needle.
Anyone interested in volunteering for this non-invasive process can participate by going to the Cambridge, Vocalis, or Carnegie Mellon sites. The more robust the voice samples they have, the faster they can develop an effective app.
C. Brown et al. 2020. Exploring Automatic Diagnosis of COVID-19 from Crowdsourced Respiratory Sound Data. In Proceedings of the 26th ACM SIGKDD International Conference on Knowledge Discovery & Data Mining (KDD '20). Association for Computing Machinery, New York, NY, USA, 3474–3484.
T. Quatieri, T. Talkar and J. Palmer, 2020. A Framework for Biomarkers of COVID-19 Based on Coordination of Speech-Production Subsystems. IEEE Open Journal of Engineering in Medicine and Biology. 203-206.
VocalisCheck Covid-19 Biomarker Whitepaper. 2020. Retrieved 11/27/20.