Bite-Weight Estimation Using Commercial Ear Buds.

Area of Science:

• Biomedical Engineering
• Human-Computer Interaction
• Digital Health

Background:

• Automatic physical activity tracking is common, but automatic eating behavior measurement is limited.
• Current commercial solutions often require specialized hardware or smartphone apps.
• There's a need for unobtrusive methods to measure food intake automatically.

Purpose of the Study:

• To evaluate the feasibility of estimating food weight per bite using only audio signals from commercial earbuds.
• To explore various audio and non-audio features with different machine learning models.
• To assess the accuracy of this approach for dietary intake monitoring.

Main Methods:

• Utilized audio signals from Samsung Galaxy Buds to capture eating sounds.
• Extracted a combination of audio and non-audio features.
• Trained and evaluated linear regression, support vector regression, and neural network models.
• Tested on an in-house dataset comprising 8 participants and 4 food types.

Main Results:

• Achieved a mean absolute error (MAE) of less than 1 gram for 3 out of 4 food types with food-specific models.
• Obtained an MAE of 2.1 grams when training a single model on all food types.
• Demonstrated improved performance compared to existing literature approaches.

Conclusions:

• Audio signals from commercial earbuds show significant potential for estimating food weight per bite.
• This method offers a low-cost, sensor-free approach to automatic dietary intake measurement.
• Further research could refine models for broader food types and real-world applications.