main.m

clc
clear
close all

addpath(genpath('./lib'));

%% parameter setting

audio_dir = './data/clean/clean_speech.wav';
% audio_dir = './data/example/SNR103F3MIC021001_ch01.wav';

mode = 2;           % 0 : ACAM3, 1 : bDNN, 2 : DNN, 3 : LSTM
threshold = 0.4;    % threshold for hard decision
output_type = 1;    % 0 : frame based prediction, 1: sample based prediction
is_default = 1;     % 0 : use trained model, 1: use default model

%% prediction
% result : binary decision
% pp : posterior probability

[result, pp] = vad_func(audio_dir, mode, threshold, output_type, is_default);

%% plot (sample based)

label_dir = './data/clean/clean_label.mat'; % groud truh directory
load(label_dir);
s = audioread(audio_dir);

figure
t = (1:length(s))./16000;
p1 = plot(t, s);
hold on
p2 = plot(t, y_label*0.3, 'g--') ;
p3 = plot(t(1:length(result)), result*0.15, 'r');
ylim([-0.3 0.6]);
xlim([0 t(end)]);
legend([p2 p3],'ground truth', 'prediction')