Model-based Bayesian signal processing in acoustics

Abstract

Bayesian signal processing has been increasingly applied to a wide variety of acoustical research and engineering tasks. Bayesian probability theory provides acousticians with an elegant framework for inferential data analysis which facilitates learning from acoustic experimental investigations that provide an improved understanding of the underlying theory. In these inferential analysis tasks, certain prior knowledge is often available about the acoustical phenomena under investigation, based either on the underlying physical theory or on certain phenomenological relationships. Bayesian probability theory allows this available information to be incorporated in the processing and analysis and exploited in the Bayesian framework as physical or phenomenological models. Many analysis tasks in acoustics often include two levels of inference, the model selection and the parameter estimation. Bayesian signal processing provides solutions to these two levels of inference by extensively using Bayes’ theorem within this unified framework. This talk will discuss various model-based approaches recently applied to signal processing and analysis in acoustics using either one or both levels of inference.