While the lockdown imposed in multiple countries is aiming to flatten the curve in the spread of the coronavirus, rapid testing remains the only way to detect and quarantine COVID-19 cases and stop the Novel coronavirus from spreading further. For now, test kits are still not available in adequate quantities around the world. But to help with early detection, AI has come to the doctor’s aid.
Researchers from Carnegie Mellon University have developed an AI-based experimental app that screens a human’s cough to detect if he or she may have COVID-19. The app is called COVID Voice Detector and is being distributed for free. The researchers claim it’s a collaboration between CMU researchers and voice scientists and researchers who work on “voice forensic technologies” from telling.ai, hat-ai.com and voca.ai.
Here’s how the app works - When you log in, there’s a tutorial for vocal prompts that includes coughing three times, reciting the alphabet and holding out vowels for as long as you can to check your lung capacity. It takes around five minutes and in the end, you see a score within 1 to 10 telling how likely your “voice carries signatures of COVID-19”.
The researchers basically match voice samples against those of known COVID-19 patients. There are options to state your demography like your age, sex, height and weight as well as if you have been diagnosed with COVID-19 or if you have any of the symptoms.
While it certainly helps a layperson get wind of a possible infection, this isn’t a replacement for a medically approved test kit. The site itself puts out this disclaimer and the fact that it is still under active development. It’s also not comparable to a COVID-19 test that’s administered by a doctor.
So while a high rating in the app may be a prompt to go get tested, that itself isn’t confirmation of COVID-19 infection since it’s based on small sample sizes. The researchers do intend to improve the app as more people, both healthy and infected, take the voice-based test.
from Latest Technology News https://ift.tt/2VbeIqr