Robust Multimodal Cough and Speech Detection using Wearables: A Preliminary Analysis

Cough detection is a crucial tool for long-term monitoring of respiratory illnesses. While clinical methods are accurate, they are not available in a home-based setting. In contrast, wearable devices offer a more accessible alternative, but face challenges in ensuring user speech privacy and detecting coughs accurately in real-world settings due to potential poor audio quality and background noise. This study addresses these challenges by developing a small-size multimodal cough and speech detection system, enhanced with an Out-of-Distribution (OOD) detection algorithm. Through our analyses, we demonstrate that combining transfer learning, a multimodal approach, and OOD detection techniques significantly improves system performance. Without OOD inputs, the system shows high accuracies of 92.59% in the in-subject setting and 90.79% in the cross-subject setting. With OOD inputs, it still maintains overall accuracies of 91.97% and 90.31% in these respective settings by incorporating OOD detection, despite the number of OOD inputs being twice that of In-Distribution (ID) inputs. This research are promising towards a more efficient, user-friendly cough and speech detection method suitable for wearable devices.