Using speakers as sensors: Detecting acoustic loads with dense neural networks and impedance features

The use of speakers as sensors to detect ear canal conditions has been previously demonstrated using variable syringe lengths attached to earphones. Building on this foundation, our research explores the potential of a single speaker to function as both an actuator and sensor by leveraging electrical impedance measurements under varying acoustic loads, analyzed with dense neural networks (DNN). Electrical impedance data, including magnitude and phase, were collected from four speakers across fourteen distinct acoustic load conditions, yielding a dataset of 5600 samples. The raw data were processed through the DNN model, achieving an 87% accuracy in length prediction independent of speaker types, which improved to 91% when incorporating speaker-specific characteristics. This innovative approach demonstrates the viability of using electrical impedance as machine learning features, paving the way for smart, sensor-integrated systems for ear health monitoring and broader biomedical applications.