SuperCollider CLASSES (extension)

SpectralEntropy
ExtensionExtension

Spectral feature extraction
 

Description

Spectral Entropy, with a choice of number of sub-bands. If one band, a measure of general peakiness of the spectral distribution.

See: SPECTRAL ENTROPY AS SPEECH FEATURES FOR SPEECH RECOGNITION Aik Ming Toh, Roberto Togneri and Sven Nordholm http://www.ee.uwa.edu.au/~roberto/research/theses/tr05-01.pdf

Pass an FFT in.

Class Methods

*kr (fft, fftsize: 2048, numbands: 1)

Arguments:

fft

input fft chain, that is, from an FFT UGen

fftsize

Size of FFT buffer must be known in advance for pre-calculation

numbands

Number of sub-bands for entropy calculation; spectral bins are collected in sub-bands, and the number of outputs of the UGen is numbands

Inherited class methods

Instance Methods

Inherited instance methods

Examples

(
{

var in, fft, entropy;

//in = SinOsc.ar(MouseX.kr(100,1000),0,0.1);
//in = Mix(SinOsc.ar([440,MouseX.kr(440,880)],0,0.1));  
in= SoundIn.ar; 

fft = FFT(LocalBuf(2048), in);

entropy=SpectralEntropy.kr(fft,2048,1);    //one output band (so full spectrum's entropy)

entropy.poll; 

Out.ar(0,Pan2.ar(0.1*Blip.ar(100,10*(entropy.sqrt)))); 
}.play
)




(
{

var in, fft, entropy;
var amplitude; 

in= SoundIn.ar; 

amplitude = Amplitude.kr(in); 

fft = FFT(LocalBuf(1024), in);

entropy=SpectralEntropy.kr(fft,1024,10);//10 bands

entropy = entropy * (amplitude.min(0.2)*5.0); //scale by amplitude to avoid low amplitude noise issues

entropy.poll; 

//Out.ar(0,Pan2.ar(0.1*Blip.ar((entropy[0])*200,entropy[1].sqrt))); 

Out.ar(0,Splay.ar(0.1*Blip.ar((entropy)*200,entropy.sqrt)));
}.play
)