Abstract

For humans, the most important functionality of auditory scene analysis is to acquire information about our everyday environments: a car approaching from behind, warning beeps, door knocking, door opening and closing and so on. Until now, most of the research on computational audio analysis has been done in the context of speech and music processing. Research on automatic detection and recognition of sound events has mostly been limited to isolated sound events, specific environments (such as meeting rooms), and small number of specific types of events.
Computational audio analysis in everyday environments has applications in areas such as multimedia content analysis, context-aware devices, assistive technologies, and acoustic monitoring. The research in the field has so far approached the task by studying two problems. The first one is context recognition, where a recording is classified into one of predefined contexts. Such contexts may be characterized by locations such as home, office, street, or grocery store or by physical and social activities. Second, sound event or acoustic event detection aims at estimating the start and end times of individual events in a recording as well as estimating a class label for each detected event. Automatic detection and recognition of events in realistic environments requires addressing many problems (eg, robustness) that have already been faced in the context of speech and music recognition. Additionally, operating with realistic sounds in everyday environments raises many new questions. Sound events can originate from very different sources and have therefore diverse acoustic …

Date

2014

Authors

Meinard Müller, Shrikanth S Narayanan, Björn Schuller

Journal

Computational Audio Analysis

Pages

11