Speakers—Used as sensors for detecting acoustic loads with artificial intelligence

Abstract

Previous research demonstrated the potential of using speakers as sensors to detect ear canal conditions. This study continues that effort by using a single speaker to measure electrical impedance across various acoustic loads. Electrical impedance data were collected and preprocessed for machine learning model training. Different image forms were tested, including magnitude only and combined magnitude phase. Using 2100 data samples with convolutional neural network-based models (AlexNet, ResNet, and DenseNet), binary and multiclass classifications achieved average accuracies of 0.9716 and 0.907, respectively. This innovative approach is set to revolutionize acoustic sensing through artificial intelligence.