Voss 1/f noise algo help

[gbreed at cix.compulink.co.uk]

kevin parks wrote:

> Hi I have a couple of attempts below at the beginnings of a 1/f^2
> Noise generator. It is in a sort of heavily commented psuedo Python
> (somewhat translated from PASCAL) You can see that there are some
> areas where i am in way over my head.  I may have introduced a mistake
> or two as well. Anyone familiar with Voss's algorithym want to help
> smooth out the rough edges here? It sure would be nice to be able to
> generate pink noise values in Python (brownian noise would be good
> too, I am trying  to work something up on that too).

I'm not familiar with this Voss algorithm, but I did write code to 
generate fractional noise in C++ a while back.  It can be a lot simpler 
than what you're writing.  Here's the important function, converted into 
pseudo-Python:

def fractionalNoise(beta, nPoints)
  """return noise with frequency spectrum 1/f**beta or something"""
  randomGenerator = Gauss()
 
  # initialize in the frequency domain
  data = []
  for i in range(nPoints):
    amplitude = ((i+1)**(-0.5*beta))*gauss()
    phase = 2*math.pi*rand()
    data.append((amplitude*math.cos(phase), amplidude*math.sin(phase)))
  
  return FFT(data)

You need three magic functions:

gauss() returns a random number according to a Gaussian distribution with 
a standard deviation of 1.  random.gauss looks suspiciously relevant

rand() is the usual random number between 0 and 1.  Must be possible.

FFT() is a fast fourier transform.  There should be one in either Numeric 
or Scientific Python.  I'm not sure exactly what the result means, but 
both real and complex parts sound like noise.

I think the standard libary includes routines for writing .wav files.  
Happy hunting!


                   Graham

      <http://www.microtonal.co.uk/>