Automatic classification of environmental noise sources by statistical methods

This paper shows how the tools of statistical pattern recognition can be used to create an "intelligent" noise monitoring system able to distinguish between the acoustic signals of different types of environmental noise sources, e.g., airplanes, cars, or trucks. The basics of statistical pattern recognition theory are reviewed in a tutorial fashion and illustrated on simple acoustical noise recognition examples. Guidelines for the design, training, and utilization of a pattern classification system are provided. The pitfalls of an intuitive, heuristic approach to these tasks are highlighted. Experimental results obtained by an automatic noise recognition system based on a statistical pattern recognition framework are presented. Finally, possible improvements of this system which are currently being investigated further are discussed.