A Two-Level Sound Classification Platform for Environmental Monitoring

Abstract EN

STORM is an ongoing European research project that aims at developing an integrated platform for monitoring, protecting, and managing cultural heritage sites through technical and organizational innovation.

Part of the scheduled preventive actions for the protection of cultural heritage is the development of wireless acoustic sensor networks (WASNs) that will be used for assessing the impact of human-generated activities as well as for monitoring potentially hazardous environmental phenomena.

Collected sound samples will be forwarded to a central server where they will be automatically classified in a hierarchical manner; anthropogenic and environmental activity will be monitored, and stakeholders will be alarmed in the case of potential malevolent behavior or natural phenomena like excess rainfall, fire, gale, high tides, and waves.

Herein, we present an integrated platform that includes sound sample denoising using wavelets, feature extraction from sound samples, Gaussian mixture modeling of these features, and a powerful two-layer neural network for automatic classification.

We contribute to previous work by extending the proposed classification platform to perform low-level classification too, i.e., classify sounds to further subclasses that include airplane, car, and pistol sounds for the anthropogenic sound class; bird, dog, and snake sounds for the biophysical sound class; and fire, waterfall, and gale for the geophysical sound class.

Classification results exhibit outstanding classification accuracy in both high-level and low-level classification thus demonstrating the feasibility of the proposed approach.