Abstract
In this paper, we present SoQr, a sensor that can be attached to an external surface of a household item to estimate the amount of content inside it. The sensor consists of a speaker and a microphone. It outputs a short duration sine wave probing sound to excite a container and its content, and then records the container’s impulse response. SoQr then extracts Mean Mel-Frequency Cepstral Coefficients from impulse response recordings of a container with different content levels and learns a support vector machine classifier. Results from a 10-fold cross validation of the prediction models on 19 common household items demonstrate that SoQr can correctly estimate the content level for these products with an average overall F-Measure above 0.96. We then further evaluated SoQr’s robustness in different usage scenarios to gain an understanding of how the system performs and specific challenges that might arise when users interact with these products and the sensor.