%demo_llado2022 - Demonstration of the use of the llado2022 model
%
%
% This demo shows how to train an own neural network to obtain localization
% estimates based on various conditions, subjective data, network parameters
% in the context of sound localization, or even for other purposes.
%
% Figure 1: Estimated direction (left) and position uncertainty (right) in the horizontal plane
%
%
% Requirements:
% -------------
%
% This demo requires the Deep Learning Toolbox. If not available,
% pre-calculated data will be downloaded and shown.
%
% See also: llado2022 exp_llado2022 demo_llado2022
%
% References:
% Lladó, Pedro, Hyvärinen, Petteri, and Pulkki, Ville. Auditory
% model-based estimation of the effect of head-worn devices on frontal
% horizontal localisation. Acta Acust., 6:1, 2022. [1]http ]
%
% References
%
% 1. https://doi.org/10.1051/aacus/2021056
%
%
% Url: http://amtoolbox.org/amt-1.5.0/doc/demos/demo_llado2022.php
% #Requirement: NNET
% #Author: Pedro Llado (2022)
% #Author: Petteri Hyvärinen (2022)
% #Author: Ville Pulkki (2022)
% #Author: Michael Mihocic (2023): if Deep Learning Toolbox not available, try loading data from cache, otherwise a warning is displayed
% This file is licensed unter the GNU General Public License (GPL) either
% version 3 of the license, or any later version as published by the Free Software
% Foundation. Details of the GPLv3 can be found in the AMT directory "licences" and
% at <https://www.gnu.org/licenses/gpl-3.0.html>.
% You can redistribute this file and/or modify it under the terms of the GPLv3.
% This file is distributed without any warranty; without even the implied warranty
% of merchantability or fitness for a particular purpose.
%% Run binaural auditory model to extract ITD and ILD from HRIR
% HRIR sofa folder
%mySofaFolder = 'SOFAfolder/';
train_lateral_angles=[0,10,30,50,70,290,310,330,350];
aux_angle = find(train_lateral_angles > 180);
angle_id = train_lateral_angles;
angle_id(aux_angle) = train_lateral_angles(aux_angle) - 360;
angle_id = sort(angle_id);
%Generate stimulus
fs = 48000;
stim = pinknoise(0.25*fs);
% Extract IRs from sofa files
%mySofaFiles = dir(mySofaFolder+"/*.sofa");
mySofaFiles(1).name = '00-OpenEars.sofa';
mySofaFiles(2).name = 'A-AKGk702-DIY.sofa';
mySofaFiles(3).name = 'B-AKGk702-Original.sofa';
mySofaFiles(4).name = 'C1-SavoxNoiseCom200-Default.sofa';
mySofaFiles(5).name = 'C2-SavoxNoiseCom200-Balanced.sofa';
%mySofaFiles = dir(mySofaFolder+"/*.sofa");
mySofaFiles(6).name = 'D-PeltorProTacII.sofa';
mySofaFiles(7).name = 'E1-BullardMagma-VisorDown.sofa';
mySofaFiles(8).name = 'E2-BullardMagma-VisorUp.sofa';
mySofaFiles(9).name = 'F-GeckoMK11.sofa';
nDevices = length(mySofaFiles);
for file_id = 1:nDevices
%sofaFile_name = strcat(mySofaFolder, mySofaFiles(file_id).name);
sofaFile_name = mySofaFiles(file_id).name;
ir_sofa = llado2022_extractirs(sofaFile_name,train_lateral_angles,fs);
if file_id == 1
ir = ir_sofa;
else
ir = [ir;ir_sofa];
end
end
% Extract binaural features (ITD and ILD)
[x_train] = llado2022_binauralfeats(ir,stim,fs);
% X_binaural_feats = binauralAuditoryModel(mySofaFolder,...
% train_lateral_angles,stim,fs,1);
% Load subjective test data
y = amt_load('llado2022', 'Y_listeningTest_labels.mat');
Y_listeningTest_labels = y.Y_listeningTest_labels;
% Col1 : device
% Col2 : sound source lateral angle
% Col3 : perceived lateral angle (average)
% Col4 : position uncertainty (average)
y_train = Y_listeningTest_labels(:,3:4);
sofaFile_name_test = 'exampleTest.sofa';
test_ir = llado2022_extractirs(sofaFile_name_test,train_lateral_angles,fs);
x_test = llado2022_binauralfeats(test_ir,stim,fs);
% NN variables
hiddenLayerSize_dir = 22;
hiddenLayerSize_uncertainty = 16;
augmentation_ratio = 10;
SNR = 60;
nIter = 2; % number of iterations to train,evaluate the model (results are averaged)
[flags,kv]=amt_configuration; % get installed toolboxes
if kv.nnet == 1
%% do full calculations if Deep Learning Toolbox available
for iter = 1:nIter
%% Train NN: direction
[net_dir,tr_dir] = llado2022_trainnn(x_train',y_train(:,1),hiddenLayerSize_dir,augmentation_ratio,SNR);
%% Train NN: uncertainty
[net_uncertainty,tr_uncertainty] = llado2022_trainnn(x_train',y_train(:,2),hiddenLayerSize_uncertainty,augmentation_ratio,SNR);
%% Evaluate a new HWD
y_est_dir(iter,:) = net_dir(x_test);
y_est_uncertainty(iter,:) = net_uncertainty(x_test);
end
if (isvector(y_est_dir) == 1 )
y_est_dir = y_est_dir;
y_est_uncertainty = y_est_uncertainty;
else
y_est_dir = mean(y_est_dir);
y_est_uncertainty = mean(y_est_uncertainty);
end
% amt_cache('set','y_est_dir',y_est_dir); % because of problems with mat2doc, amt_cache does not work in demos
% amt_cache('set','y_est_uncertainty',y_est_uncertainty); % because of problems with mat2doc, amt_cache does not work in demos
else
%% get precalculated data to plot figure if Deep Learning Toolbox not available
% y_est_dir = amt_cache('get','y_est_dir'); % because of problems with mat2doc, amt_cache does not work in demos
% y_est_uncertainty = amt_cache('get','y_est_uncertainty'); % because of problems with mat2doc, amt_cache does not work in demos
x=amt_load('llado2022','demo_cache.mat');
y_est_dir=x.y_est_dir;
y_est_uncertainty=x.y_est_uncertainty;
end
if isempty(y_est_dir) || isempty(y_est_uncertainty)
% data could neither be calculated nor loaded from cache
error('Data could neither be calculated (Deep Learning Toolbox not found), nor loaded from auxdata. Figure cannot be created.');
else
% plot figure
plot_llado2022(y_est_dir,y_est_uncertainty,angle_id);
end