Smart wearables get smarter at catching coughs
Researchers have unveiled a significant improvement in cough detection for wearable health monitors, enabling more reliable tracking of chronic conditions and timely health interventions. By combining audio signals with movement data collected directly from chest-worn devices, the new approach sharpens the ability to distinguish coughing from ordinary speech and other nonverbal sounds. This refinement promises to make cough frequency a more dependable biomarker for monitoring respiratory health and predicting events such as asthma attacks.
The challenge of detecting coughs in the real world
Wearable cough detection has long faced a daunting hurdle: sounds of coughing can closely resemble speech, sneezes, throat-clearing, or even groans. “In real-world settings, models encounter sounds they have never heard before, which degrades their accuracy,” explains Edgar Lobaton, a professor of electrical and computer engineering at North Carolina State University and the paper’s corresponding author. Traditional models were trained on labeled libraries of sounds, but when deployed, their performance often dropped due to the variability of everyday acoustic environments.
A multimodal solution: audio plus movement
The researchers addressed this problem by leveraging two types of data collected from chest-worn health monitors. First, they captured audio data representing the sounds around the wearer. Second, they gathered data from the device’s accelerometer, which records movement. The key insight is that coughing generates a distinctive combination of sound and sudden bodily movement, and using both signals helps disambiguate coughs from other actions that produce similar movement alone.
As Yuhan Chen, the study’s first author and a recent NC State Ph.D. graduate, notes, “Movement alone cannot reliably identify coughing, because activities like laughing or talking can create similar motion patterns. But when we fuse sound with movement, we gain complementary information that improves cough detection accuracy.”
<h2 Multimodal data, refined algorithms
The team did not rely solely on raw data fusion. They built on prior work to refine the underlying algorithms, training models on real-world readings from chest-mounted devices. Importantly, the data set included both naturally occurring coughs and other sounds captured in everyday life, helping the model learn to separate coughing from noncough events more effectively.
<h2 Laboratory results point to practical benefits
In controlled laboratory evaluations, the multimodal model demonstrated superior accuracy compared with earlier approaches. Notably, it produced fewer false positives, meaning fewer instances where the system mislabeled noncough noises as coughs. While the team acknowledges that no model is perfect and room for improvement remains, the results mark a meaningful advance toward dependable cough monitoring in real-world wearable use.
<h2 Implications for patient care and research
Accurate cough detection has wide-ranging clinical implications. Cough frequency can serve as a valuable biomarker for tracking the progress of respiratory diseases and for predicting when a patient might experience an asthma exacerbation or require an inhaler. For researchers, reliable cough data from wearables offers a noninvasive means to quantify symptom dynamics over time, supporting personalized treatment plans and proactive interventions.
<h2 About the study
The work, titled “Robust Multimodal Cough Detection with Optimized Out-of-Distribution Detection for Wearables,” was published in the IEEE Journal of Biomedical and Health Informatics. The research team included Feiya Xiang, Alper Bozkurt, Michelle Hernandez, and Delesha Carpenter, from NC State and UNC-Chapel Hill, among others. The study was supported by the National Science Foundation and NC State’s Center for Advanced Self-Powered Systems of Integrated Sensors and Technologies (ASSIST).
<h2 Looking ahead
Researchers are continuing to refine the approach to tackle remaining challenges, such as handling a broader range of coughing styles and environmental conditions. The ultimate goal is to deploy highly reliable cough-detection capabilities in consumer wearables, enabling users, clinicians, and researchers to monitor respiratory health passively and continuously.
