SymRecorder: Detecting Respiratory Symptoms in Multiple Indoor Environments Using Earphone-Microphones

Respiratory symptoms associated with sound frequently manifest in our daily lives. Despite their potential connection to illness or allergies, these symptoms are often overlooked. Current detection methods either depend on specific sensors that must be deliberately worn by the user or are sensitive to environmental noise, limiting their applicability to specific settings. Considering that indoor environments vary, we propose SymRecorder, an earphone microphone-based application, for detecting respiratory symptoms across a range of indoor settings. By continuously recording audio data through the earphone’s built-in microphone, we can detect the four common respiratory symptoms: cough, sneeze, throat-clearing, and sniffle. We have developed a modified ABSE-based method to detect respiratory symptoms in noisy environments and mitigate the impact of noise. Additionally, a Hilbert transform-based method is employed to segment the continuous respiratory symptoms that users may experience. Based on selected acoustic features, the four symptoms are classified using the residual network and the multi-layer perceptron. We have implemented SymRecorder on various Android devices and evaluated its performance in multiple indoor environments. The evaluation results demonstrate SymRecorder’s dependable ability to detect and identify users’ respiratory symptoms in various indoor environments, achieving an average accuracy of 92.17% and an average precision of 90.04%.