Sound monitoring networks new style

Abstract

The research project IDEA: intelligent distributed environmental assessment, is aimed at using new capabilities offered by consumer hardware to deploy sound monitoring methodologies that approach human environmental sound perception as closely as possible. The low cost noise measurement nodes deployed in the network pre-process the sound to a data set that can be transmitted continuously to the central servers. On these servers basic acoustic features are extracted and a self organizing map is trained to represent commonly co-occuring ones. Once an initial mapping is available all incoming sound is analysed and recorded when (1) the combination of features is not recognized, indicating unexpected sounds; (2) the feature set matches so closely a unit in the map that the sound can be used as a prototype. Unrecognized sounds trigger attention which leads to further training of the self organized map and potential alerts. The implementation of the above methodology resulted in a tool that allows getting detailed and useful information on the sounds that people will observe in a certain environment by mapping frequency of occurrence - and dominance of different classes of sounds that can easily be auralised for example in the context of soundscape design.