demo_vggvox_verif_voxceleb2.m

function demo_vggvox_verif_voxceleb2(varargin)
% DEMO_VGGVOX_VERIF - minimal demo with the VGGVox model pretrained on the
% VoxCeleb2 dataset for Speaker Verification

% downloads the VGGVox model and
% prints the distance on a test evalutation pair

opts.modelPath = '' ;
opts.gpu = 1;
opts.dataDir = 'testfiles/verif'; 
opts = vl_argparse(opts, varargin) ;

% Example speech segments for input
inpPath1 = fullfile(opts.dataDir, '8jEAjG6SegY_0000008.wav');
inpPath2 = fullfile(opts.dataDir, 'x6uYqmx31kE_0000001.wav'); 

% Load or download the VGGVox model for Verification pretrained on VoxCeleb2
modelName = 'ver_net.mat' ;
paths = {opts.modelPath, ...
    modelName, ...
    fullfile(vl_rootnn, 'data', 'models-import', modelName)} ;
ok = find(cellfun(@(x) exist(x, 'file'), paths), 1) ;

if isempty(ok)
    fprintf('Downloading the VGGVox model for Verification ... this may take a while\n') ;
    opts.modelPath = fullfile(vl_rootnn, 'data/models-import', modelName) ;
    mkdir(fileparts(opts.modelPath)) ; base = 'http://www.robots.ox.ac.uk' ;
    url = sprintf('%s/~vgg/data/voxceleb2/%s', base, modelName) ;
    urlwrite(url, opts.modelPath) ;
else
    opts.modelPath = paths{ok} ;
end
load(opts.modelPath); net = dagnn.DagNN.loadobj(netStruct);



%% Audio settings. 
 opt.audio.window   = [0 1];
 opt.audio.fs       = 16000;
 opt.audio.Tw       = 25;
 opt.audio.Ts       = 10;            % analysis frame shift (ms)
 opt.audio.alpha    = 0.97;          % preemphasis coefficient
 opt.audio.R        = [];  % frequency range to consider
 opt.audio.M        = 40;            % number of filterbank channels
 opt.audio.C        = [];            % number of cepstral coefficients
 opt.audio.L        = [];            % cepstral sine lifter parameter%keyboard;


net.meta = opt; 

% add a distance layer 
net.addLayer('dist', dagnn.PDist('p',2), {'x0_s1', 'x0_s2'}, 'distance');

% Evaluate network on GPU and set it to test mode
net.move('gpu');
net.conserveMemory = 0;
net.mode = 'test' ;

% Setup buckets to allow for average pooling 
buckets.pool 	= [2 5 8 11 14 17 20 23 27 30];
buckets.width 	= [100 200 300 400 500 600 700 800 900 1000];

% Load input pair and do a forward pass
inp1 = test_getinput(inpPath1, net.meta, buckets);
inp2 = test_getinput(inpPath2, net.meta, buckets);

s1 = size(inp1,2);
s2 = size(inp2,2);

p1 = buckets.pool(s1==buckets.width);
p2 = buckets.pool(s2==buckets.width);

ind1 = net.getLayerIndex('pool_time_b1'); 
ind2 = net.getLayerIndex('pool_time_b2'); 
net.layers(ind1).block.poolSize=[1 p1];
net.layers(ind2).block.poolSize=[1 p2];

featid = strcmp({net.vars.name},'distance');
net.eval({ 'data_b1', gpuArray(inp1) ,'data_b2', gpuArray(inp2) });
dist = gather(squeeze(net.vars(featid).value));

% Print distance
fprintf('dist: %05d \n',dist); % should output a small distance if the two segments come from the same identity