抄録
Tracking exactly what, when, and how much people drink in everyday life is crucial for health care. Still, existing methods, such as containermounted capacity sensors and camera-based beverage classifiers, only address a single aspect. On the other hand, container-mounted sensors have limited real-world applications due to the need to measure the container. Here, we present a small wearable system that fuses an earbud microphone with a throat microphone to recognize beverage type and volume simultaneously. In a controlled study, our approach was applied to 10 beverage types (apple juice, coffee, cola, energy drink, green tea, orange juice, protein drink, soda, tomato juice, and water), classified with an F1-Score of 0.56. Volume was measured with an average absolute error of less than 3.89 mL. These results demonstrate the potential for continuous and highly accurate fluid and nutrient intake monitoring in everyday settings.
https://ror.org/01hvx5h04 
