Since the start of the pandemicwe discovered PCR and antigen tests, the principle of which is to push a swab deep into our nose. Neither glam nor pleasant, we would have done without it. But so far it was the one and only way to know whether or not we were infected with the Covid-19. Today, it would seem that the voice is opening up to other methods of virus search. And this, thanks to artificial intelligence. The idea? Detect the virus thanks to… our voice. It was a team of researchers from the University of Maastricht who came up with the idea by analyzing data from the application. COVID-19 Sounds, developed by the University of Cambridge.
We’re building a dataset of respiratory sounds to help combat #Covid19. You can help! We need as many people as possible to download our app and contribute!👉 Please share: https://t.co/yr2hJV9WwD
Data is anonymous & only used for #research. @Cambridge_CL @Cambridge_Uni
— Covid-19 Sounds App (@Covid19Sounds) June 11, 2020
89% reliable results
The researchers had access to 893 audio samples from 4,352 people, of whom 308 had tested positive. The users of the application had to answer a questionnaire concerning their health and then had to record breathing sounds. Thus, they had to breathe deeply through their mouth 3 to 5 times, cough three times, and read a short sentence three times. For understand and analyze this datathe researchers used a voice analysis method, Mel Spectogram, which identifies different characteristics such as volume, power and fluctuations.
Visit our section Health for more news!
This new Wikipedia page on “Apps to analyze #Covid_19 sounds » features our @Cambridge_Uni study👉https://t.co/EDQLBMsa8e
Don’t forget you can contribute and help research! @Cambridge_CL #COVID2019 pic.twitter.com/bXlJE545mO
— Covid-19 Sounds App (@Covid19Sounds) May 21, 2020
“In order to distinguish the voice of COVID-19 patients of those who did not have the disease, we built different models of artificial intelligence and assessed which one worked best for classifying COVID-19 cases” explained Wafaa Aljbawi, researcher member of the project. This model ofartificial intelligence is 89% accurate for detecting a positive case and 83% for a negative case.
” These results promising suggest that simple voice recordings and fine-tuned AI algorithms can potentially achieve a high precision to determine which patients are infected with COVID-19. »explains Wafaa Aljbawi. “Such tests can be provided free and are simple to interpret. Additionally, they enable remote virtual testing and have a turnaround time of less than a minute. They could be used at entry points to large gatherings, allowing rapid screening. » Promising results but which also highlight the interest of using AI in the service of health.