Detection of Fluid Intake Swallowing Events Using Acoustic Signals and Template Matching

Swallowing is a fundamental physiological process for nutrition intake. Monitoring and analysing swallowing activities have been a pivotal research topic due to their role in maintaining human health. This paper presents a template-matching approach to detect swallowing events by analysing sound signals captured through a throat microphone. The microphone was positioned unobtrusively around the neck, ensuring minimal interference with natural swallowing processes. By matching the unique acoustic characteristics associated with swallowing with the pre-established template of swallowing signals, this approach achieves high accuracy in distinguishing swallowing from speech, coughing and other noises. A total number of 363 swallowing data, 140 non-swallowing data and 180 background (null) data were collected from 7 subjects. The swallowing and non-swallowing classification accuracy reached an F1-score of 0.95, outperforming conventional machine learning techniques. For swallowing detection from a signal stream, the system achieved an F1-score of 0.88, indicating that this method has the potential for non-invasive and real-time fluid intake monitoring.