“Does the speech of a person infected with COVID-19 show unique vocal sounds, features and patterns — and do they change as the illness progresses?”

That’s the hypothesis of a new research study led by Allegheny Health Network (AHN) and Pittsburgh-based voice diagnostic startup Telling.ai.

What the study discovers could change how COVID-19 and other novel coronaviruses are detected and monitored in the future.

COVID-19 causes severe acute respiratory distress and progression in varied and unpredictable ways. Some patients experience mild symptoms followed by a period of wellness, and then the sudden appearance of severe respiratory symptoms.

Anil Singh, system director for pulmonary/critical care/allergy-immunology and sleep medicine at AHN, and his collaborators at Telling.ai believe voice analysis could provide clues for treatment.

Patterns might show if a COVID-19 infection is about to progress to a severe illness that may require hospitalization. They may even detect COVID-19 infection in asymptomatic patients.

All monitoring could be conducted remotely, reducing the virus’ spread.

“For physicians treating COVID-19 infection, one of the most difficult questions has been determining when a patient needs hospitalization,” Singh said. “We are hopeful that voice analysis may provide valuable information that could help doctors recognize oncoming distress and quickly get patients the treatment they need, potentially preventing critical illness.”

Singh’s partnership with Telling.ai began before the coronavirus pandemic. Collaborative studies looked at the possibility of voice recordings replacing lengthy lung function tests in people with chronic obstructive pulmonary disease (COPD).

Initial results show a near 1-to-1 correlation between results of the traditional lung function test and testing done by analyzing voice recordings.

The emergence of COVID-19 redirected this research. Singh and other researchers are now exploring how voice analysis might play a role in improving detection and treatment of coronavirus.

“The ability to deliver instantaneous, accurate insights about respiratory health over the phone has the potential to save lives,” said Raj Dhaka, Telling.ai co-founder and CEO. “By donating their voice samples, patients suspected of having COVID-19 may help us open a new front in the war to detect and treat the infection.”

Dhaka said Telling.ai’s algorithms show a strong correlation between voice and “widely accepted lung function metrics.”

The study is open to AHN patients who are receiving a physician-ordered test for COVID-19. They can provide voice samples using a website

If they subsequently test positive for COVID-19, they will report their symptoms and provide additional voice samples over a 30-day period.

Researchers will examine how changes in the voice samples correlate with patients’ symptoms, visits to an emergency department or physician appointments. The results will not impact the care they receive.

If the results are promising, Singh and Telling.ai will design a clinical trial for FDA approval, opening the way for broader use of the technology.