Most of us are aware that, with the ongoing COVID-19 global pandemic, a lot of our scientists and medical experts strive to formulate a cure. While there are some ways developed to at least diagnose signs before the virus even spreads, sometimes it takes a lot of our time and money to participate.

After seeing cheap and accurate diagnostic measures for COVID-19, a graduate student from Carnegie Mellon University, Benjamin Striner, suggested a more convenient solution that all of us can access.

Together with a group of researchers from the university and other institutions, he released an early version of an app, which they claimed could diagnose COVID-19 by analyzing our voice.

It may sound ambitious considering the degree of seriousness of this pandemic; still, Striner believes that the COVID Voice Detector’s algorithm could be of great value to track the virus’ spread. The Carnegie Mellon University has put up a page on their website where you can submit your voice sample.

Carnegie Mellon University COVID19 Voice Detector Sample Submission - THESIS.PH

Limitations and Important Things to Consider About the App

A professor from the same university, Bhiksha Raj, discussed that they are working on developing a voice-based solution made possible by initial experiments and prior expertise.

However, Raj noted that the results from the app are untested. It is preliminary and, according to her, the scores only show how much signatures in our voice matches those of COVID patients they tested.

Moreover, some of the algorithm’s limitations suggest that it is:

  • still highly experimental
  • a work in progress
  • not approved by the CDC or FDA
  • not accurate as laboratory tests
  • should not be a basis for making healthcare decisions and treated as medical advice

Given the COVID-19-related problems that we face worldwide, the app may indeed be useful in tracking and handling possible outbreaks. It is very convenient as it can easily monitor us when we access it through our smartphones or computers with microphones.

The process of collecting our data is also simple, as it only asks us to:

  • cough many times
  • recite the alphabet
  • record vowel sounds

After so, the app presents score as a progress bar that reports your likelihood to have COVID-19.

Rita Singh, also a professor at the university, states that a COVID patient’s cough is distinctive. Given her expertise in identifying micro-signatures in our voice, she explained that COVID affects our breathing patterns and several vital parameters so severely that it gives strong signatures in our voices.

The team worked with colleagues around the world since their campus is closed due to the pandemic. To strengthen the evidence for their app’s specification and teach the algorithm to spot differences, they gathered audio from patients who are:

  • healthy
  • have COVID-19 virus
  • acquired other viruses

Moreover, a professor from Columbia University Medical Center, Ashwin Vasan, added that this could only be a distraction from things that are necessary, such as PPEs and test kits. However, Striner expressed adherence to having more false positives than false negatives, saying that perhaps some people get tested without really needing it.

The app’s accuracy is difficult to quantify at the moment, but their efforts all aim at refining the existing algorithm.

This app’s limitations are your opportunities to innovate it. Do you plan to develop a similar app for your thesis?


Please enter your comment!
Please enter your name here