Classification of meeting-room acoustic events with support vector machines and variable-feature-set clustering

Acoustic events produced in meeting-room-like environments may carry information useful for perceptually aware interfaces. We focus on the problem of classifying 16 types of acoustic events, using and comparing several types of features and various classifiers based on either GMM or SVM. A variable-feature-set clustering scheme is developed and compared with an already reported binary tree scheme. In our experiments with event-level features, the proposed clustering scheme with SVM achieves a 31.5% relative error reduction with respect to the best result from a binary tree scheme.