MIT researchers analyzed cough recordings from thousands of volunteers—including those with and without COVID-19, some asymptomatic—to develop an AI tool for early virus detection and pandemic control.
Detecting asymptomatic carriers poses a major challenge, as they show no symptoms. A study published in the IEEE Journal of Engineering in Medicine and Biology on September 28, 2020, introduces an innovative screening method. MIT experts found that coughs from infected individuals differ subtly from those of healthy people.
Leveraging artificial intelligence, the model processes cough audio captured via smartphone or computer microphones. Human ears can't distinguish these differences, but the AI—built on proven neural networks for neurobiological disease detection—identifies lung performance issues and vocal cord changes linked to the virus.
“Speech and cough sounds are both influenced by the vocal cords and surrounding organs. This means that when you speak, part of your conversation is similar to coughing, and vice versa. It also means that things we easily guess from fluent speech, such as the person's gender, native language, or even emotional state, can be detected by coughing by the AI,” explains Brian Subirana, co-author of the study.
The team trained the model on tens of thousands of cough samples and 1,000 hours of speech data. Results show 98.5% accuracy for symptomatic COVID-19 coughs and 100% for asymptomatic carriers.
The vision: a free, user-friendly mobile app for widespread screening. With approval from governments and health authorities, it could enable anytime detection of asymptomatics, curbing transmission—especially before entering public spaces.
