Overview: A Leap Forward in Cough Detection for Wearables
Researchers have unveiled a new approach to detecting coughs with wearable health monitors that combine audio data and accelerometer movement. This multimodal strategy improves the accuracy of cough detection, a vital capability for monitoring chronic respiratory conditions and predicting events such as asthma exacerbations. By better distinguishing coughs from speech and nonverbal sounds, wearables can offer more reliable insights for patients and clinicians alike.
Why Cough Detection Matters
Coughs serve as an important biomarker for a wide range of health conditions. Tracking cough frequency helps monitor the progression of respiratory diseases and can indicate when an asthma patient might need intervention, such as using an inhaler. As a result, there is strong interest in technologies that can detect and track coughs in real time, especially in everyday settings outside of clinics.
The Challenge of Real-World Sound Recognition
Despite advances, cough-detection models have struggled to separate coughing sounds from speech, sneezes, throat-clearing, groans, and other everyday noises. This difficulty is amplified when models encounter unfamiliar sounds in real life. “Cough-detection models are trained on a library of sounds, but when the model runs across a new sound, its ability to distinguish cough from not-cough suffers,” explains Edgar Lobaton, a professor of electrical and computer engineering at NC State and corresponding author of the study.
The Multimodal Solution: Audio plus Movement
To tackle this challenge, the researchers tapped two data streams from chest-worn health monitors: audio captured by the device and movement data from an accelerometer. The combination of sound and movement provides complementary information. Movement patterns alone cannot reliably indicate coughing—different actions can produce similar motions—but when paired with audio cues, coughs can be identified with higher confidence. “Movement provides context that supports sound-based detection,” notes Yuhan Chen, the paper’s first author.
Real-World Data and Improved Algorithms
Beyond real-world sound samples, the team built on prior work to refine the underlying algorithms. Their laboratory tests showed a notable reduction in false positives—instances where non-cough sounds were misclassified as coughs. This improvement means clinicians and patients can trust the readings more, making cough frequency a more dependable metric for monitoring disease activity and therapy effectiveness.
Implications for Patients and Clinicians
The work represents a meaningful step toward more reliable, wearable-based health monitoring. If deployed broadly, multimodal cough detection could help patients manage conditions like asthma more proactively, guiding decisions about when to use rescue inhalers and when to seek medical attention. Clinicians could also leverage aggregated cough data to track disease trends over time and tailor treatment plans accordingly.
About the Study
The research, titled “Robust Multimodal Cough Detection with Optimized Out-of-Distribution Detection for Wearables,” appears in the IEEE Journal of Biomedical and Health Informatics. It was led by NC State researchers and collaborated with UNC and other contributors, with support from the National Science Foundation and NC State’s ASSIST program.
Looking Ahead
While the results are promising, the authors acknowledge there is room for further improvement. Future work will focus on enhancing robustness to new environments and refining models to further minimize false positives while maintaining sensitivity. The team is optimistic about bringing these improvements into consumer-friendly wearables, where real-time cough monitoring can support smarter, data-driven health management.
