Source: MachineListening.sc

Given an FFT **chain** this calculates the *Spectral Flatness* measure, defined as a power spectrum's geometric mean divided by its arithmetic mean. This gives a measure which ranges from approx 0 for a pure sinusoid, to approx 1 for white noise.

The measure is calculated linearly. For some applications you may wish to convert the value to a decibel scale - an example of such conversion is shown below.

buffer |
an FFT chain. |

s.boot; ( { // Example - vary mixture of white noise and pure tone with the mouse var in, chain, flat, flatdb, flatdbsquish; in = XFade2.ar(WhiteNoise.ar, SinOsc.ar, MouseX.kr(-1,1)); chain = FFT(LocalBuf(1, 2048), in); Out.ar(0, in * 0.1); flat = SpecFlatness.kr(chain); flatdb = 10 * flat.log; // Convert to decibels flatdbsquish = LinLin.kr(flatdb, -45, -1.6, 0, 1).max(-10); // Rescale db roughly to 0...1. flat.poll(10, "flatness: "); flatdb.poll(10, "flatness (db): "); [flat, flatdbsquish] }.scope; ) ( { // Now try with your own voice var in, chain; in = SoundIn.ar([0,1]).mean; chain = FFT(LocalBuf(1, 2048), in); Out.kr(0, [in, SpecFlatness.kr(chain).poll(1, "flatness: ")]); }.scope; )

helpfile source: /usr/local/share/SuperCollider/HelpSource/Classes/SpecFlatness.schelp

link::Classes/SpecFlatness::

link::Classes/SpecFlatness::