// load a source
b = Buffer.read(s,"/Volumes/machins/projets/newsfeed/sons/textes/Audio/synth/fromtexttospeech-AmE-George.wav")
b.play

//slightly oversegment with novelty
//segments should still make sense but might cut a few elements in 2 or 3
~originalslices = Buffer(s);
FluidBufNoveltySlice.process(s, b, indices: ~originalslices, feature: 1, kernelSize: 29, threshold: 0.05, filterSize: 5, hopSize: 128, action: {~originalslices.numFrames.postln;})

//test the segmentation by looping them
(
{
	BufRd.ar(1, b,
		Phasor.ar(0,1,
			BufRd.kr(1, ~originalslices,
				MouseX.kr(0, BufFrames.kr(~originalslices) - 1), 0, 1),
			BufRd.kr(1, ~originalslices,
				MouseX.kr(1, BufFrames.kr(~originalslices)), 0, 1),
			BufRd.kr(1,~originalslices,
				MouseX.kr(0, BufFrames.kr(~originalslices) - 1), 0, 1)), 0, 1);
}.play;
)

//analyse each segment with MFCCs in a dataset
~originalslices.getn(0,~originalslices.numFrames, {|x|~originalslicesarray = x; if ((x.last != b.numFrames), {~originalslicesarray = ~originalslicesarray ++ (b.numFrames)}); });//retrieve the indices and add the file boundary at the end if not there already

//iterates through the
//a few buffers and our dataset - with back and forth from the language
(
~mfccs = Buffer(s);
~stats = Buffer(s);
~flat = Buffer(s);
~slices = FluidDataSet(s,\slices);

Routine{
	s.sync;
	(~originalslicesarray.size - 1).do{|i|
		FluidBufMFCC.process(s, b, startFrame: ~originalslicesarray[i], numFrames: (~originalslicesarray[i+1] - ~originalslicesarray[i]), numChans: 1,features: ~mfccs, numCoeffs: 20, action: {
			FluidBufStats.process(s, ~mfccs, startChan: 1, stats: ~stats, action: {
				FluidBufFlatten.process(s, ~stats, ~flat, action: {
					~slices.addPoint(i.asSymbol, ~flat);
				});
			});
		});
	};
}.play;
)

~slices.print
~slices.clear

//run a window over consecutive segments, forcing them in 2 classes, and merging the consecutive segments of similar class
//we overlap the analysis with the last (original) slice to check for continuity
(
~winSize = 6;//the number of consecutive items to split in 2 classes;
~query = FluidDataSetQuery(s);
~kmeans = FluidKMeans(s,2,100);

Routine{
	~indices = [0];
	~head = 0;
	~windowDS = FluidDataSet(s,\windowDS);
	~windowLS = FluidLabelSet(s,\windowLS);
	~sliceDict = Dictionary;
	~tempDict = Dictionary.new;

	s.sync;

	~slices.dump{|x|~sliceDict = x;};
	s.sync;

	while ( {~head <= (~originalslicesarray.size - ~winSize)},
		{
			var step = ~winSize - 1;
			//run a process on ~winSize items from ~head (with an overlap of 1)
			//copy the items to a subdataset
			~winSize.do{|i|
				~tempDict.put((i.asString), ~sliceDict["data"][(i+~head).asString]);//here one could curate which stats to take
			};
			~windowDS.load(Dictionary.newFrom([\cols, 133, \data, ~tempDict]));
			s.sync;

			//kmeans 2 and retrieve ordered array of class assignations
			~kmeans.fitPredict(~windowDS,~windowLS, {
				~windowLS.dump{|x|~assignments = x.at("data").atAll(x.at("data").keys.asArray.sort{|a,b|a.asInteger < b.asInteger}).flatten;};
			});
			s.sync;
			~assignments.postln;

			step.do{|i|
				if (~assignments[i+1] != ~assignments[i], {~indices= ~indices ++ (~originalslicesarray[~head+i+1])});
			};

			~head = ~head + step;
	});

	//leftovers
	if ( (~originalslicesarray.size - ~head) > 1, {
		//run a process on (a.size - ~head) items from ~head
		(~originalslicesarray.size - ~head - 1).do{|i|
				if (~assignments[i+1] != ~assignments[i], {~indices= ~indices ++ (~originalslicesarray[~head+i+1])});
			// (~head+i).postln;
		};
	});
	~indices.postln;
}.play;
)

{var i = 8;BufRd.ar(1,b,Line.ar(~originalslicesarray[i],~originalslicesarray[i+1],(~originalslicesarray[i+1] - ~originalslicesarray[i])/b.sampleRate, doneAction: 2))}.play;
{var i = 4;BufRd.ar(1,b,Line.ar(~indices[i],~indices[i+1],(~indices[i+1] - ~indices[i])/b.sampleRate, doneAction: 2))}.play;