BARUMERLI2023
Bayesian spherical sound localization model (multi-feature)

Usage:

[m] = barumerli2023('sofa_obj', sofa_obj);
[m] = barumerli2023('template', template, 'target', target);

Input parameters:

`sofa_obj`	SOFA structure containing listener HRTFs.
`template`	Templates from barumerli2023_featureextraction.
`target`	Targets from barumerli2023_featureextraction.

Output parameters:

m

Matrix as in localizationerror with the actual and predicted directions. barumerli2023_metrics can be used for further analysis of m.

doa

Structure with the fields:

estimations: Matrix. Size: (target_num x num_exp x pos) with pos being coordinates in the Cartesian coordinate system.
posterior: Posterior distribution. Size: (target_num x template_num x num_exp).

coords

coord Object with the actual directions of the binaural sounds.

Description:

barumerli2023(...) is an ideal-observer model of human sound localization, by which we mean a model that performs optimal information processing within a Bayesian context. The model considers all available spatial information contained within the acoustic signals encoded by each ear. Parameters for the optimal Bayesian model are determined based on psychoacoustic discrimination experiments on interaural time difference and sound intensity.

Additional input parameters:

`'num_exp'`	Number of repetitions. For each target repeat the experiment num_exp times. Default: 50
`'sigma_itd'`	Standard deviation associated to noise added to the ITD feature. Default: 0.569
`'sigma_ild'`	Standard deviation associated to noise added to the ILD feature. Default: 1
`'sigma_spectral'`	Standard deviation associated to noise added to the spectral features. Default: 1.25
`'sigma_prior'`	Standard deviation of the prior distribution. If set to empty [], a uniform distribution is considered. Default:11.5
`'sigma_motor'`	Standard deviation of the motor noise. If set to empty [], motor noise is disabled. Default: 14

If no SOFA object is provided then the model requires:

`'template'`	internal representation specified by a specific feature space. Refer to barumerli2023_featureextraction for its computation.
`'target'`	preprocessed target struct with the binaural sounds of which the direction has to be estimated. Refer to barumerli2023_featureextraction for its computation.

Further, cache flags (see amt_cache) can be specified.

References:

R. Barumerli, P. Majdak, M. Geronazzo, D. Meijer, F. Avanzini, and R. Baumgartner. A Bayesian model for human directional localization of broadband static sound sources. Acta Acust., 7:12, 2023. [ DOI ]

THE AUDITORY MODELING TOOLBOX