Acoustic Event Classification for Ambient Assisted Living and Healthcare Environments

Authors: Hussein Hussein, Marc Ritter, Robert Manthey, Jan Schloßhauer, Etienne Fabian, Manuel Heinzig

Abstract:

Acoustic events that are produced by people can be used to recognize activities or other critical behavior. This contribution presents our first experiments on acoustic event classification for utilization in the sector of healthcare. Ten acoustic events, including speech and non-speech events, which are usually occurred in this field are defined. The database of acoustic events is collected in a recording studio and annotated manually. A variety of features and several classifiers have been proposed for classification of acoustic events in order to detect the best feature set and classifiers for the specified acoustic events. Low-level audio features and the corresponding delta features are utilized. Statistical functionals are applied to each of the features and delta features. The best obtained classificaton results, calculated by the F-Measure, for the ten aoustic events with a feature set of 430 features is 92.50%.