Beyond loudness

Abstract

Low-cost sensor networks have increasingly been used to monitor noise pollution as an alternative to certified sound level meters. Existing sensor networks monitor loudness and, in some cases, classify sound sources. Most sensors do not capture metrics that are more representative of the human perception of sound, require a permanent power supply, or are relatively expensive. We develop an energy-efficient soundscape sensor with the goal of recording metrics complementary to loudness. We have implemented metrics from psychoacoustics and metrics inspired by biodiversity research, such as sharpness, intermittency, and acoustic entropy. Furthermore, the sensor predicts the source of sound events. For privacy-preservation, all audio is processed directly on the sensor. The sensor is based on a low-power microcontroller (ESP32-S3), available for a fraction of the cost of a Raspberry Pi, which is often used for sound source prediction. Finally, the sensor is solar-powered and therefore easy to install for research purposes at places without direct access to the power grid. A temporary deployment of several sensors in Amsterdam, the Netherlands, is planned.