Another interesting source of info on this subject is a paragraph in the book "Elements of Computer Music" by F.R. Moore, a classic in Computer Music. There the generation method for 1/f^alpha noise is "spectrum shaping"(1), which at the end is not very different from the Kasdin method implemented in cnoise patch
(1): noise -> fft -> "shape" the spectrum by multiplication on the desired function (1/x, 1/x^2 etc.) -> ifft -> normalize in audio range (+1, -1)
Alberto
On Mon, 21 Aug 2006, alberto.zin@poste.it wrote:
if I'm not wrong, as long as it is white it is flat in
spectral terms
but the distribution of amplitudes (probability density
function, pdf)
of white noise can be gaussian, cauchy, uniform, etc.
Therefore I'm not
sure that the spectrum is proportional to the pdf..
Sorry, my mistake. I was completely confused. The spectrum that we're talking about is actually the expectation of a random spectrum, which is the expectation of amplitude of each frequency separately, which doesn't say much about which pdf each frequency has.
Btw, the spectrum of white noise is not flat, it's just that the expectation of the spectrum is flat.
What does "uniform" mean in a situation where things are understood in terms of amplitude and phase? are the amplitudes selected in polar coordinates with the correct jacobian? if instead cos and sin components are selected uniformly, then the maximum amplitude will only be possible for four phases, which are the four corners of a square shape.
_____________________ ... | Mathieu Bouchard - tél:+1.514.383.3801 - http://artengine.ca/matju | Freelance Digital Arts Engineer, Montréal QC Canada