Online Acoustic Event Detection

Abstract: The detection and classification of acoustic events is an important task with many practical applications like video understanding, surveillance or speech enhancement.
In this talk a brief overview of applications and some methods for online acoustic event detection will be given. A Bag-of-Features approach for acoustic event detection will be introduced and discussed in detail. In the Bag-of-Features principle a feature representation is learned
from the training data. These methods proved successful in many pattern recognition tasks like
document analysis, gesture recognition and especially image classification.
There are multiple extensions to the of Bag-of-Features methods which are improving the capabilities. Here, lessons that can be learned from the Computer Vision domain will be outlined showing some positive and negative examples.
At the end of the talk, an outlook on open challenges in Acoustic Event Detection will be given. These include multi sensor fusion, overlapping events and dataset/labeling issues.