Computational analysis of acoustic events in everyday environments.

Sounds carry a large amount of information about our everyday environment and physical events that take place in it. Recent advances in machine learning allows automatic methods to analyze this information, for example, by detecting and classifying acoustic events produced by various sources. This allows several applications, for example, in acoustic surveillance, context-aware devices, and multimedia indexing. This talk will present signal processing and machine learning methods that can be used to detect and classify everyday acoustic events originating, e.g., from vehicles, human activity, human and animal vocalizations, in everyday environments. It will describe the scientific challenges in such methods, for example, many sources having highly similar spectral characteristics and multiple sources being active simultaneously. It will explain how state-of-the-art methods based on advanced deep neural network topologies deal with these challenges. The talk will also discuss the practical challenges related to the development of the methods, such as acquisition of data that is used to develop the methods. It will present results from recent evaluations of event detection systems and illustrate them using audio and video examples.