THE AUDITORY MODELING TOOLBOX

Applies to version: 1.1.0

View the help

Go to function

BARUMERLI2021 - Bayesian spherical sound localization model (multi-feature)

Program code:

function [varargout] = barumerli2021(varargin)
%BARUMERLI2021  Bayesian spherical sound localization model (multi-feature)
%   Usage: [results,template,target] = barumerli2021(template,target);
%
%   Input parameters:
%       template:           internal representation specified by a specific feature space
%
%
%   BARUMERLI2021 accepts the following optional parameters:
%
%       target          : preprocessed target struct
%
%
%   Further, cache flags (see amt_cache) can be specified.
%
%   Description: 
%   ------------
%
%   BARUMERLI2021(...) 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.
%
%
%
%   See also: barumerli2021_featureextraction
%   barumerli2021_metrics
%
%   Url: http://amtoolbox.org/amt-1.1.0/doc/models/barumerli2021.php

% Copyright (C) 2009-2021 Piotr Majdak, Clara Hollomey, and the AMT team.
% This file is part of Auditory Modeling Toolbox (AMT) version 1.1.0
%
% This program is free software: you can redistribute it and/or modify
% it under the terms of the GNU General Public License as published by
% the Free Software Foundation, either version 3 of the License, or
% (at your option) any later version.
%
% This program is distributed in the hope that it will be useful,
% but WITHOUT ANY WARRANTY; without even the implied warranty of
% MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
% GNU General Public License for more details.
%
% You should have received a copy of the GNU General Public License
% along with this program.  If not, see <http://www.gnu.org/licenses/>.

%   References: barumerli2021

%   #StatusDoc: Good
%   #StatusCode: Submitted
%   #Verification: Unknown
%   #Requirements: MATLAB SOFA M-STATISTICS M-Control M-Signal
%   #Author: Roberto Barumerli, Information Engineering Dept., University of Padova, Italy, 2020



%% Check input options
definput.keyvals.num_exp = 1;

definput.keyvals.sofa_obj = struct([]);
definput.keyvals.template = struct([]);
definput.keyvals.target = struct([]);
definput.keyvals.sigma_lateral =  0.569;
definput.keyvals.sigma_lateral_bis = 1;
definput.keyvals.sigma_polar = [];
definput.keyvals.prior_vector = [];
definput.keyvals.prior_data = [];
definput.keyvals.sigma_prior = 11.5;

definput.keyvals.sigma_sensorimotor = [];

[flags, kv]  = ltfatarghelper({}, ...
                             definput, varargin);

if ~isempty(kv.sofa_obj)
    %% prepare feature space
    [template, target] = barumerli2021_featureextraction(kv.sofa_obj);
elseif ~isempty(kv.template) && ~isempty(kv.target)  
    template = kv.template;
    target = kv.target;
else
    error(['Please specify a SOFA object or the ', ...
             'template and the targets. See barumerli2021_featureextraction.'])
end

template_feat = [template.lateral, template.lateral_bis, template.polar];
target_feat = [target.lateral, target.lateral_bis, target.polar];
assert(~isempty(template.lateral_bis))

if ~isempty(template.polar) 
    if numel(kv.sigma_polar) == 1
        polar_len = size(template.polar, 2);
        lateral_bis_len = size(template.lateral_bis, 2);
        sigma = blkdiag(kv.sigma_lateral^2*eye(size(template.lateral, 2)), ...
                        kv.sigma_lateral_bis^2*eye(lateral_bis_len), ...
                        (kv.sigma_polar)^2*eye(polar_len));
    elseif(numel(kv.sigma_polar) == size(template.polar,2))
        assert(numel(kv.sigma_polar) == size(template.polar,2), ...
            'problem with dimensions with sigma_polar and template_polar')
        lateral_bis_len = size(template.lateral_bis, 2);
        sigma = blkdiag(kv.sigma_lateral.^2*eye(size(template.lateral, 2)), ...
                        kv.sigma_lateral_bis.^2*eye(lateral_bis_len), ...
                        diag(kv.sigma_polar));
    elseif sqrt(numel(kv.sigma_polar)) == size(template.polar,2) % the user specified the covariance matrix for the polar dimension
        lateral_bis_len = size(template.lateral_bis, 2);
        sigma = blkdiag(kv.sigma_lateral.^2*eye(size(template.lateral, 2)), ...
        kv.sigma_lateral_bis.^2*eye(lateral_bis_len), ...
        kv.sigma_polar);
    else
        error('Something went wrong with the definition of the polar uncertainty.')
    end
else
    sigma = blkdiag(kv.sigma_lateral^2*eye(size(template.lateral, 2)), ...
                    (kv.sigma_lateral_bis)^2*eye(size(template.lateral_bis, 2)));
end

sigma_inv = inv(sigma);

%% prior computation 
if ~isempty(kv.prior_vector)
    kv.prior_vector = kv.prior_vector(:);
    
    assert(abs(sum(kv.prior_vector) - 1) < 1e-10, ...
                'Prior vector does not sum to one')
    
    if size(kv.prior_vector,1) == size(template.coords.pos,1)
        prior = kv.prior_vector;
    else
        error('given prior vector must have the same size of template.coords')
    end
elseif ~isempty(kv.sigma_prior)
    sph = template.coords.return_positions('spherical');

    prior = exp(-0.5*(sph(:,2)/kv.sigma_prior).^2);  
    prior = prior./sum(prior);
elseif ~isempty(kv.prior_data)
    assert(size(kv.prior_data, 2) == 8, ['The 2nd dimension of the', ...
        ' prior_data matrix needs to be equal to 8', ...
        ' (plz check localizationerror.m)']);
    [prior, dirs] = prior_compute(kv.prior_data);
    
    for tc = 1:size(template.coords.pos, 2)
        assert(sum((dirs(:,tc) - template.coords.pos(:,tc)).^2) < eps, ...
            'prior vector is not aligned with template directions') 
    end
else
    prior = ones(1,length(template.coords.pos));
end
%% internal belief computation
template_num = size(template_feat, 1);
target_num = size(target_feat, 1);
doa_MAP = zeros(target_num, kv.num_exp);
doa_estimations = zeros(target_num, kv.num_exp, 3);
lik_log = zeros(target_num, template_num, kv.num_exp);
temp_c = template.coords.return_positions('cartesian');
temp_hp = template.coords.return_positions('horizontal-polar');

for e = 1:kv.num_exp
    for ta = 1:target_num % number of targets
        likelihood = zeros(template_num, 1);

        %% AWGN NOISE
        x = mvnrnd(target_feat(ta,:),sigma);

        %% COMPUTE POSTERIOR
        for te = 1:template_num % number of directions in the template
            u_diff = (x-template_feat(te,:));
            posterior(te, 1) = (exp(-0.5*u_diff*sigma_inv*transpose(u_diff)))*prior(te);            
        end

        % normalize
        posterior = (posterior./sum(posterior)); 
 
        %% decision stage
        [~, doa_MAP(ta,e)] = max(posterior);
        lik_log(ta,:,e) = posterior;
        doa_estimations(ta,e,:) = temp_c(doa_MAP(ta,e),:);
        
    end
end

%% SENSORIMOTOR SCATTERING
if ~isempty(kv.sigma_sensorimotor)
    for e = 1:kv.num_exp
        doa_estimations(:,e,:) = sensorimotor_scatter_von_mises(squeeze(doa_estimations(:,e,:)), ...
            kv.sigma_sensorimotor);
    end
end

% results 
doa.estimations = doa_estimations;
doa.likelihood = lik_log;

% return m matrix 
varargout{1} = barumerli2021_metrics(doa, target.coords, 'm');

% return cartesian estimations and other stuff
if nargout > 1
    varargout{2} = doa;
end

% return cartesian actual directions
if nargout == 3
    varargout{3} = target.coords;
end


function dirs_new = sensorimotor_scatter_von_mises(dirs, sigma_m)
% add sensorimotor noise to localization estimations
%
%%
% Input parameters:
%   m                   localization matrix (check localizationerror.m)
%   sigma               standard deviation in degrees
%   
% Output paramenters
%   m_new     localization matrix with scattered estimations
%   dirs      directions in cartesian coordinates
%
% AUTHOR: Roberto Barumerli
% Information Engineering Dept., University of Padova, Italy, 2020

    assert(size(dirs, 2) == 3 | numel(dirs) == 3)
    assert(sigma_m >= 5, 'sensorimotor concentration can lead to complex values')
    % handle if size(dirs) = [3,1]
    if numel(dirs) == 3
        if size(dirs, 1) == 3
            dirs = dirs';
        end
    end

    dirs_new = zeros(size(dirs));

    kappa = 1/deg2rad(sigma_m)^2;

    for i=1:size(dirs,1)
        dirs_new(i,:) = Random_vMF_Inv(kappa, dirs(i,:), 1);
    end

function Y = Random_vMF_Inv(kappa,varargin) % MODIFIED BY ROBARU
% RANDOM_VMF_INV Generates a random sample of size n from the 
%                von Mises-Fisher distribution
%
% Y = Random_vMF_Inv(kappa,Mu,n)
%%
% Examples of correct usage:
%
% n=2^12; 
% kappa = 4.2;
%
% Mu=[0 0 1];% Mu=[1 -1 1] ;%  Mu=[0 -1 1]; 
%
% Examples of correct function construction: 
% Y = Random_vMF_Inv(kappa,Mu,n)
% Y = Random_vMF_Inv(kappa,Mu)
% Y = Random_vMF_Inv(kappa)
%%
%
% Required Inputs
% kappa       distribution parameter (kappa \in {Real Numbers})
%
% Optional Inputs
% Mu          distribution parameter, (for rotating the North pole to Mu)
% n             sample size ( where n is a positive integer)
%
% Authors: Gy.Terdik, B.Wainwright
%
% Copyright 2018 Gy.Terdik
%
% Licensed under the Apache License, Version 2.0 (the "License");
% you may not use this file except in compliance with the License.
% You may obtain a copy of the License at
%
% http://www.apache.org/licenses/LICENSE-2.0
%
% Unless required by applicable law or agreed to in writing, software
% distributed under the License is distributed on an "AS IS" BASIS,
% WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
% See the License for the specific language governing permissions and
% limitations under the License.
%
%% Verify/Default parameter values

    % Required parameter characteristics
    p = inputParser;
    dbl = {'double'};
    class = {'numeric'};
    pos_int = {'integer','positive'};
    real = {'real'};
    noinf = {'finite'};

    % Check characteristics for the required and optional parameters
    addRequired(p,'kappa',@(x)validateattributes(x,dbl,real));
    addOptional(p,'Mu',[],@(x)validateattributes(x,dbl,noinf));
    addOptional(p,'n',[],@(x)validateattributes(x,class,pos_int));

    p.parse(kappa,varargin{:});
    Mu = p.Results.Mu;
    n = p.Results.n;

    % Default Values
    if isempty(Mu)
        Mu = [0 0 1];
    end
    if isempty(n)
        n = 2^15;
    end

    %% density
    % dim of the space / m=3 corresponds to the unit sphere
    if kappa == 0;
        Y = Random_Uni_Norm(n); % kappa=0 corresponds to uniform dist. on the sphere
        return
    else
        X = vMF_Inv(kappa,n); % cos(theta)
    end

    %%
    psi = 2*pi*rand(n,1);
    sX = sqrt(1-X.^2); 
    Y = [cos(psi).*sX,sin(psi).*sX,X];

    %% rotation (orient mean vector (mu) with the North Pole)
    Np=[0,0,1]; % z-axis (North Pole)
    Mu=Mu/norm(Mu); % should be unit vector
    if norm(Mu-Np) > eps % !!!!!modified!!!!!!!!!!!
        Y=Rotation_Mu(Y,Mu);
    end


function X=vMF_Inv(kappa,n)
% VMFB Generates a random sample of size N from the von Mises-Fisher distribution
%
% W1=vMFb(kappa,N)
%
% kappa         distribution parameter, concentration parameter (kappa >= 0)
% N             sample size ( where N is a positive integer)
%
% Authors: Gy.Terdik, B.Wainwright
%
% Copyright 2018 Gy.Terdik
%
% Licensed under the Apache License, Version 2.0 (the "License");
% you may not use this file except in compliance with the License.
% You may obtain a copy of the License at
%
% http://www.apache.org/licenses/LICENSE-2.0
%
% Unless required by applicable law or agreed to in writing, software
% distributed under the License is distributed on an "AS IS" BASIS,
% WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
% See the License for the specific language governing permissions and
% limitations under the License.
%
%
%
% The methods implemented here are based on
% "Simulation of the von mises fisher distribution"
% by Andrew T.A. Wood (1994)
% Communications in Statistics - Simulation and Computation, 23:1, 157-164,
% DOI: 10.1080/03610919408813161
% dim of the space / m=3 corresponds to the unit sphere

%% Rubinstein 81, p.39, Fisher 87, p.59

     kappaS=sign(kappa);
     kappa=abs(kappa);
%      kappa1=exp(-2*kappa);
     U = rand(n,1); %random n-by-1 vector from uniform(0,1)
     X=log(2*U*sinh(kappa)+exp(-kappa))/kappa;
%      cos(2*asin(sqrt(-log(U*(1-kappa1)+kappa1)/(2*kappa)))); %

    X=kappaS*X(1:n);
    
function Y=Rotation_Mu(X,Mu)
% ROTATION_MU rotates the vector of points so that it coincides with the
%             North Pole (a.k.a. the standard z-axis)

%% rotation (orient mean vector (mu) with the North Pole)
Np=[0,0,1]; % z-axis (North Pole)
Mu=Mu/norm(Mu); % should be unit vector
    
if Mu(3)~= 1
    Ux=cross(Np,Mu);%axis of rotation
    Ux= Ux/norm(Ux);
    thetaX=acos(Mu(3));
    Rg= rotationVectorToMatrix(Ux*thetaX);
    Y=X*Rg;
end