An environment aware ML estimation of acoustic radiation pattern with distributed microphone pairs

This paper presents a parametric approach to classify the radiation pattern of an acoustic source given the signals captured by multiple microphones. The radiation pattern influences the way the acoustic waves propagate within an enclosure, with direct implications on the behavior of most audio processing algorithms. In particular, the Generalized Cross-Correlation PHAse Transform is affected by the emission pattern as well as by the orientation of the source. A Maximum Likelihood estimator is introduced by using descriptors of the acoustic characteristics of the environment, e.g. wall absorption coefficients and room dimensions, from which models of the observed Generalized Cross-Correlation PHAse Transform are derived for a specific emission pattern. A generic unimodal source directivity is modeled using a parameterized cardioid function. A sub-band implementation is proposed to account for the frequency dependence of the source emission pattern. Experiments on simulated and real data show that the acoustic radiation pattern can be estimated in an effective way under noisy and reverberant conditions.