A probabilistic framework for subband autoregressive models applied to room acoustics

Real room acoustic impulse responses (AIRs) modelled by infinite impulse response (IIR) filters require high model orders. Many problems involving the estimation of AIRs reduce to high dimensional optimisation problems. Subband autoregressive (AR) modelling techniques reduce this difficult optimisation problem to a number of simpler low dimensional optimisations. This paper introduces a formulation for subband AR modelling in a probabilistic framework which facilitates robust Bayesian parameter estimation. The paper also provides new results to show that the subband AR representation accurately models typical AIRs and, therefore, is suitable for modelling room reverberation