A recent study introduced a soft skin-attachable throat vibration sensor (STVS) designed for real-time monitoring of throat-related events. This device accurately tracks activities like coughing, speaking, swallowing, and throat clearing, achieving a classification accuracy of 95.96%.
The STVS was engineered to adhere to the curved skin of the neck, enabling continuous measurement of vibrational signals from the larynx and pharynx. The research team developed a deep ensemble model that integrates various neural networks trained on multi-modal data, which includes time-series features and acoustic spectral image patterns.
The classification model was evaluated on a test dataset containing diverse throat-related events across multiple languages, demonstrating its effectiveness in accurately identifying these activities. The system's design allows for precise detection of subtle throat signals that conventional microphones may overlook.
In the study, 9000 data segments were gathered from 32 subjects, with each segment lasting 625 ms. The data was processed through a series of signal-processing steps, including normalization and adaptive filtering, enhancing the quality of the acoustic signals.
The STVS hardware consists of a sensing part and a controller, connected by a serpentine interconnect that ensures stable signal transmission during neck movements. The sensing part is strategically placed above the laryngeal prominence to capture high-quality vocal signals.
Results indicated that the system maintained high accuracy even in dynamic environments, showing a classification accuracy of 95.16% while subjects walked at a speed of 8 km/h. The innovative monitoring system demonstrates significant potential for improving dysphagia management through continuous and precise throat activity tracking.