Researchers believe that using AI to analyze voice could help doctors screen patients for potential coronavirus symptoms. The voice analysis test would take under five minutes and could be self-administered.
(TNS) — You know it when you hear it: a friend’s voice sounds scratchy, a parent struggles to finish a sentence without pausing to take a few breaths. Something doesn’t sound right, and maybe it’s time to see a doctor.
Doctors at Allegheny Health Network and researchers with the technology startup Telling.ai are working together to tap into that instinct by developing a voice analysis test to screen for a variety of respiratory illnesses, including covid-19.
The voice analysis test, which was developed with researchers from the Carnegie Mellon School of Computer Science and Language Technology Institute, would identify subtle changes in the voice that suggest something is wrong. Artificial intelligence tools would match those signals to a label for their condition. Doctors could then use that data to inform what type of care — for example, a trip to an emergency room, or just more time resting at home — a patient might need next.
“If there was a way to predict by just the voice where this individual could go, that could be pretty dramatic in the management of these patients prior to them ever ending up in a hospital,” said Dr. Anil Singh, system director for pulmonary care at Allegheny Health Network.
The voice analysis test would not be used to diagnose whether a patient has covid-19. It could be used to determine whether a patient has a respiratory ailment that requires a follow-up visit to a doctor or additional testing, or to monitor how a patient’s symptoms are progressing over time, Singh said.
The technology builds off research started prior to the coronavirus pandemic that used voice analysis to screen for COPD (chronic obstructive pulmonary disease).
The test takes less than five minutes and can self-administered by a patient using a cellphone, Singh said.
The patient is given several prompts, including saying the “ah” sound for as long as possible, and taking a few deep breaths. The patient is also asked to read a passage that includes all 44 phonemes in the English language — sounds that are building blocks for words, like the sound the letter C makes in “car.”
The voice analysis considers a patient’s voice and breath to listen for issues like wheezing and coughing. Listening to a person read a passage helps the technology map certain features of that person’s speech.
“By just listening to you, especially if I’ve spoken to you before, I can tell if you sound hoarse, do you have a cold,” said Satya Venneti, chief technology officer at Telling.ai. “We already make judgments about people from their voices. What we’re doing is helping machines make the same kind of judgments.”
A preliminary study that included about 130 patients found a strong correlation between the results of a pulmonary function test — the “gold standard” for diagnosing COPD, Singh said — and the results of voice analysis.
A pulmonary function test measures lung function, including how fast a patient can breathe and how much a patient can breath in and out.
Researchers are now working to expand that study to at least 500 patients, including people who are receiving a physician-ordered test for covid-19 at an Allegheny Health Network test site, Singh said.
Patients visiting Allegheny Health Network covid-19 testing centers started receiving packets about the study and explaining how they can participate last week.
If those patients test positive for covid-19, they will be asked to report their symptoms and provide additional voice samples over a 30-day period to help researchers understand how changes in the voice correlate with symptoms or visits to an emergency room.
Singh and Telling.ai will design a clinical trial for FDA approval as this study progresses.
©2020 The Tribune-Review (Greensburg, Pa.) Distributed by Tribune Content Agency, LLC.
Never miss a story with the daily Govtech Today Newsletter.