dan@0: function eventdetection(dirFuncs, dirInput,nameInput,dirOutput,nameOutput,COindex)
dan@0: %% Readme of the eventdetection function
dan@0: % Input parameters
dan@0: %       dirFuncs:   should link to the directory with the toolboxes
dan@0: %                   e.g. 'D:\Projects\AASP\functions'
dan@0: %       dirInput:   should be the directory linking to the test scripts
dan@0: %                   e.g. 'D:\Projects\AASP\Datasets'
dan@0: %       nameInput:  is the name of the test set
dan@0: %                   e.g. 'test01.wav'
dan@0: %       dirOutput:  Is the directory where the output file shoud be saved
dan@0: %                   e.g. 'D:\Projects\AASP\Output'
dan@0: %       nameOutput: Is the name of the output files
dan@0: %                   e.g. 'outputTest01.txt'.
dan@0: %       C0index:    Determines the loaded threshold
dan@0: %                   Use 1 for Office Life
dan@0: %                   Use 2 for Office Synthetic with a SNR of -6
dan@0: %                   Use 3 for Office Synthetic with a SNR of 0
dan@0: %                   Use 4 for Office Synthetic with a SNR of 6%       
dan@0: %% add paths to the functions
dan@0: addpath([dirFuncs filesep 'functions' filesep 'challange']);
dan@0: addpath([dirFuncs filesep 'functions' filesep 'fe_funcs']);
dan@0: addpath([dirFuncs filesep 'functions' filesep 'funcsMobilab']);
dan@0: %% load the GMMs
dan@0: load([pwd filesep 'acousticModel' filesep 'GMM.mat']);
dan@0: %% Load the threholds
dan@0: nrStages = 5;
dan@0: minC0 = [-189.5 -35 -45 -95];
dan@0: %% Load the development audio file
dan@0: dirDataName = [dirInput filesep nameInput];
dan@0: [x_develop fs] = wavread(dirDataName);
dan@0: disp(['Stage 1 of ' num2str(nrStages) ' is completed']);
dan@0: %% Apply a downssampling to new_fs=16kHz
dan@0: new_fs=16000;
dan@0: x_develop = resample(x_develop, new_fs, fs);
dan@0: fs = new_fs;
dan@0: %% Extract the features
dan@0: features=feature_extraction(x_develop, audioconf);
dan@0: disp(['Stage 2 of ' num2str(nrStages) ' is completed']);
dan@0: %% Determine where C0 > minimum
dan@0: C0 = features.mfcc_static(14,:);
dan@0: frames=(C0>minC0(COindex));
dan@0: %% Do moving average filtering on the indices
dan@0: minWindowFrames=50;
dan@0: % Filter coefficients
dan@0: b_mov_avg=ones(minWindowFrames,1)/minWindowFrames;
dan@0: frames_filt=fftfilt(b_mov_avg,double(frames));
dan@0: % Seek for ones and go to doubles
dan@0: frames_filt_ones=double((frames_filt>=0.999));
dan@0: % Seek risig edges for compensating with the filter delay
dan@0: % Each rising edge corresponds to an event
dan@0: b_edges=[1 -1];
dan@0: edges=fftfilt(b_edges,frames_filt_ones);
dan@0: [ignore indRisingEdges]=find(edges>=0.999);
dan@0: [ignore indFallingEdges]=find(edges<=-0.999);
dan@0:  nrEdges=size(indRisingEdges,2);
dan@0: % Preallocation of variable eventFrames for speed
dan@0: eventFrames=zeros(nrEdges,2);
dan@0:     for(edgeNr=1:nrEdges)
dan@0:         frames_filt_ones(indRisingEdges(edgeNr)-minWindowFrames+1:indRisingEdges(edgeNr))=1;
dan@0:         eventFrames(edgeNr,:)=[indRisingEdges(edgeNr)-minWindowFrames+1 indFallingEdges(edgeNr)];
dan@0:     end,
dan@0: indEvents=find(frames_filt_ones);
dan@0: indSilence=find(ones(1,length(C0))-frames_filt_ones);
dan@0: disp(['Stage 3 of ' num2str(nrStages) ' is completed']);
dan@0: %% Compare the developpment script with all the GMMs (posteriorgram)
dan@0: % Preallcoation of variables for speed
dan@0: nrClasses = size(gmm_class_mfcc_feat,1);
dan@0: likelihood = zeros(nrClasses,length(C0));
dan@0: % Loop over the classes and determine likelihood
dan@0:     for(classNr=1:nrClasses)
dan@0:         likelihood(classNr,:) = pdf(gmm_class_mfcc_feat{classNr},features.mfcc_d_dd');
dan@0:     end,
dan@0: % Compare with the silence class GMM    
dan@0: likelihood(classNr+1,:) = pdf(gmm_silence_mfcc_feat,features.mfcc_d_dd');
dan@0: labelsClass{classNr+1}='silence';
dan@0: % Go to posteriors
dan@0: posterior=bsxfun(@rdivide,likelihood,sum(likelihood,1));
dan@0: %% Apply an moving average filtering on the posteriorgram
dan@0: % Preallocation for speed
dan@0: likelihoodFilt = zeros(nrClasses,length(C0));
dan@0: % Min duration devided by 2 as filterlengths
dan@0: minDuration = [0.325 0.3599 0.3612 0.4448 0.7662 1.026 0.725 0.4601 0.5899 0.2379 0.7102 0.243 4.2318 6.1003 0.487 0.0579 0.0579]/2;
dan@0: numFrames = floor((minDuration-(audioconf.framelen_ms/1000)+(audioconf.framestep_ms/1000)) / (audioconf.framestep_ms/1000));
dan@0: % Loop over the posteriorgram
dan@0:     for(classNr=1:nrClasses+1)
dan@0:         b=(ones(numFrames(classNr),1)/numFrames(classNr));
dan@0:         likelihoodFilt(classNr,:) = fftfilt(b,posterior(classNr,:));             
dan@0:     end,
dan@0: % Make sure no complex values occur (some have a very small imaginary valuee
dan@0: likelihoodFilt = abs(likelihoodFilt);
dan@0: % Back to posteriogram
dan@0: posteriorFilt=bsxfun(@rdivide,likelihoodFilt,sum(likelihoodFilt,1));
dan@0: disp(['Stage 4 of ' num2str(nrStages) ' is completed']);
dan@0: %% Apply the thresholding on the moving averaged posteriorgram
dan@0: posteriorFiltActive=posteriorFilt;
dan@0: posteriorFiltActive(:,indSilence)=0;
dan@0: % Loop over the detected events
dan@0:     for(detectionNr=1:nrEdges)
dan@0:         [ignore(detectionNr) ID(detectionNr)] = max(mean(posteriorFilt(:,eventFrames(detectionNr,1):eventFrames(detectionNr,2)),2));
dan@0:     end,
dan@0: %% Go over to the AASP metrics    
dan@0: pianorollClassification=zeros(size(posteriorFiltActive));
dan@0: for(detectionNr=1:nrEdges)
dan@0:     pianorollClassification(ID(detectionNr),eventFrames(detectionNr,1):eventFrames(detectionNr,2))=ones;
dan@0: end,
dan@0: % Go to a text file complementary to the challange requirements
dan@0: eventBased(pianorollClassification,nameOutput,dirOutput, audioconf);
dan@0: disp(['Stage 5 of ' num2str(nrStages) ' is completed']);
dan@0: end