A method for recognition of coexisting environmental sound sources based on the Fisher’s linear discriminant classifier

A method for sound recognition of coexisting environmental noise sources by applying pattern recognition techniques is developed. The investigated technique could benefit several areas of application, such as noise impact assessment, acoustic pollution mitigation and soundscape characterization. This study distinguishes from other investigations by focusing on cases where the noise sources appear mixed (i.e., several noise sources might be present at the same time in one location), which is a more realistic and frequent situation in cities than a single sound source without other interfering noises.

The identification and, furthermore, the estimation of the contribution of each source to the overall level is one important goal in the current investigation, which would improve environmental noise assessment in complex situations. The method for recognizing the noise sources in adverse conditions is based on the Fisher’s Linear Discriminant classifier, and estimates noise source contributions based on a distance measure of vector projections.

The method is able to identify mixed sources in 96% of the 27 tested signals and to correlate the contribution of the individual sources with their sound pressure level. The results obtained from tests in real city environments show an accurate performance in the description of the sound scenarios.