Granular Synthesis


Server.default=s=Server.internal;



In Granular Synthesis sounds are modelled out of microscopic particles of sound, short grains in the region of 10-100 milliseconds long.
































There are lots of choices for these grains, they might be enveloped sine tones, or tiny extracts of sampled sound:


b = Buffer.read(s,Platform.resourceDir +/+ "sounds/a11wlk01.wav");


//three different possible grains 

(

{

var singrain1, singrain2, sfgrain;

singrain1=SinOsc.ar(440,0,XLine.kr(1.0, 0.0001, 0.05));

singrain2=FSinOsc.ar(800, 0.0, Line.kr(1.0,0,0.05).squared);

sfgrain=(PlayBuf.ar(1,b.bufnum, BufRateScale.kr(b.bufnum)))*EnvGen.kr(Env([0,1,1,0],[0.01,0.01,0.01], -4));


[singrain1,singrain2,sfgrain]

}.plot(0.1,s)

)






















When lots of these microsounds are combined into big swarms, we can make macroscopic soundscapes.  


We can control the swarms by scheduling grains over time, forming clouds of events.



//simple sine grain synthdef - note the all important doneAction

(

SynthDef(\sinegrain, {arg pan, freq, amp; var grain; 


grain= SinOsc.ar(freq, 0, amp)*(XLine.kr(1.001,0.001,0.1,doneAction:2)-0.001);


Out.ar(0,Pan2.ar(grain, pan))}).add; 

)


//listen to a single grain;

Synth(\sinegrain,[\freq,rrand(100,10000),\amp, exprand(0.05,0.1), \pan, 1.0.rand2]);


//schedule 100 random grains over 1 second 

(

{

100.do{arg i; 


Synth(\sinegrain,[\freq,rrand(100,10000),\amp, exprand(0.05,0.1), \pan, 1.0.rand2]);  

0.01.wait

}; 

}.fork

)











Over time, we can change the characteristics of the swarm, for example, playing with its density, and control individual parameters of grains within tendency masks or following certain paths


//schedule 200 random grains over time, decreasing the range of allowed random frequencies and lowering the density over time

(

{

200.do{arg i; 

var timeprop = (i/199.0)**3;


Synth(\sinegrain,[\freq,exprand(100,5000-(20*i)),\amp, exprand(0.05,0.1), \pan, 1.0.rand2]);  

rrand((timeprop*0.1).max(0.01),timeprop*0.3).wait

}; 

}.fork

)














//simple playbuf grain synthdef - note the all important doneAction

(

SynthDef(\sfgrain, {arg bufnum=0, pan=0.0, startPos=0.0, amp=0.1, dur=0.04; 

var grain; 


grain= PlayBuf.ar(1,bufnum, BufRateScale.kr(bufnum), 1, BufFrames.ir(bufnum)*startPos, 0)*(EnvGen.kr(Env.perc(0.01,dur),doneAction:2)-0.001);


Out.ar(0,Pan2.ar(grain, pan))}).add; 

)



b = Buffer.read(s,Platform.resourceDir +/+ "sounds/a11wlk01.wav");



//individual grain

Synth(\sfgrain,[\bufnum, b.bufnum, \startPos,rrand(0.0,1.0),\amp, exprand(0.005,0.1), \pan, 1.0.rand2]);  



//schedule 200 random soundfile playback grains over time, with random offset positions into the soundfile and lowering the density over time

(

{

200.do{arg i; 

var timeprop = (i/199.0)**3;


Synth(\sfgrain,[\bufnum, b.bufnum, \startPos,rrand(0.0,timeprop),\amp, exprand(0.005,0.1), \pan, 1.0.rand2]);  

rrand((timeprop*0.1).max(0.01),timeprop*0.4).wait

}; 

}.fork

)




















Each grain might have many different parameters attached to it; some salient ones might be the pitch, the duration of the envelope, the pan position in the stereo field or the amplitude. The overall cloud can also have some sort of distribution for these parameters, which might lead to a tendency mask determining the range of frequencies of the particles allowed at differnet points in time, or control of the evolving density of the cloud. 


The composer's work is to both specify the grains, and also control how they are used over time to make an interesting compositional structure. 



These techniques were first conceptualised and explored in instrumental and electronic music by Iannis Xenakis (late 50s), and further investigated in computer implementation, notably by Curtis Roads and Barry Truax, from the early 1970s on. Real-time systems became plausible in the 1980s.  



























Because you can take tiny slices of sound, granular processing allows one to perform quite dramatic transformations on sound sources. The sound can be made to disappear into a stream of tiny quanta and reappear, coalescing out of distinct particles:



(

var w, slid, lastval;

lastval=0.0;


w=Window("My Window", Rect(100,500,200,200)); 

//A 200 by 200 window appears at screen co-ordinates (100, 500)


slid=Slider(w,Rect(10,10,150,40)); //a basic slider object 


slid.action_({lastval= slid.value;}); //this is the callback- the function is called whenever you move the slider


{

inf.do{arg i; 


var prop, timestart, timeend;


prop= (i%300)/300;


timestart= prop*0.8;

timeend= prop*(0.8+(0.1*lastval));


Synth(\sfgrain,[\bufnum, b.bufnum, \startPos,rrand(timestart,timeend),\amp, exprand(0.005,0.1), \pan, lastval.rand2, \dur, 0.1+(lastval*0.5)]);  

//max in this to avoid ever going near 0.0 wait time, which would crash the computer!

(((lastval*0.2)+0.01).max(0.01)).wait

}; 

}.fork;



w.front;

)























Some classes to explore:


[PitchShift] //A granular pitchshifter


[TGrains] //efficient granular synthesis on a buffer, from a single UGen


Also see various granular UGens by Josh Parmenter added for SC3.2: GrainSin, GrainFM, GrainBuf, GrainIn, Warp1


To read more on these techniques, Curtis Roads's Computer Music Tutorial is a good introduction, and his Microsound book explores many more possibilities.