Maastricht, Netherlands - Developers of a mobile phone app based on artificial intelligence (AI) say it can reliably identify whether you have Covid-19 from the sound of your voice.
Researchers say this is possible because the virus usually hits the upper respiratory tract and vocal cords, which changes the sound of your voice.
The AI was able to detect coronavirus infections with 89% accuracy, which is better than many of the lateral flow tests available, where accuracy varies widely depending on the brand, scientists said on Monday.
Plus, lateral flow tests are even less reliable when it comes to detecting Covid-19 in people who are asymptomatic, Wafaa Aljbawi, a researcher at the Institute of Data Science, Maastricht University, The Netherlands, told a congress of respiratory specialists.
"These promising results suggest that simple voice recordings and fine-tuned AI algorithms can potentially achieve high precision in determining which patients have Covid-19 infection," she told the European Respiratory Society International Congress in Barcelona, Spain.
For the test to work, users of the app were asked to provide some recordings - coughing, breathing deeply, and reading a short sentence on screen.
The tests can be provided at no cost and are simple to interpret, she said, thereby enabling remote, virtual testing.
Plus, they have a turnaround time of less than a minute, meaning they could be used at the entry points for large gatherings, enabling rapid screening, Aljbawi said.
The app's overall accuracy was 89%. It correctly detected positive cases in 89% of cases and correctly identified negative cases 83% of times.
That makes it significantly more useful than lateral flow tests, which wrongly classify a larger number of people as negative, allowing many to spread the virus further, falsely thinking they are healthy, she said.
The app could be especially useful in poorer countries or places where highly accurate yet more expensive PCR tests are difficult to come by, the team said, calling for larger-scale studies to validate their research.
The team, led by Sami Simons, pulmonologist at Maastricht University Medical Center, and Visara Urovi, also from the Institute of Data Science, used voice recordings provided by a crowd-sourcing app run by the University of Cambridge.
They used Mel-spectrogram analysis to assess 893 audio samples from 4,352 healthy and non-healthy participants, 308 of whom had tested positive for Covid-19. The analysis measured voice features such as loudness, power, and variation over time.