This study investigates the use of speakers (earphones) as sensors for detecting acoustic loads through variations in electrical impedance. Leveraging electroacoustic coupling and machine learning-based analysis, impedance features were used to predict insertion depth with high accuracy. The dense neural network-based classification achieved 87% overall accuracy, improving to 91% with speaker-specific data and up to 99% for individual units. By identifying optimal frequency-domain regions in which impedance responses were most sensitive to changes in acoustic loading, classification performance further improved. The results were also compared with a regression model, which yielded higher error rates but indicated a promising direction for future work. This work integrated electroacoustic principles with machine learning to establish a foundation for smart, low-cost, and portable diagnostic systems for ear health and related biomedical applications.
Kim et al. (Sun,) studied this question.