Andy Farnell wrote:
Could a stats mathematician please help me check this. (attached)
Well I'm not one but...
I'm following the Box Muller formula for getting a cheap Gaussian distribution (instead of adding up 12 sources a la central limit method).
http://www.dspguru.com/howto/tech/wgn.htm
Does this look right?
Also, I have no idea how to check the distribution. It sounds the same as uniform noise and looks the same in the spectrograph? What gives? Do I need to average over a very long time or something to see any difference?
The way I understand it, white noise has the characteristic that if you count the number of occurrences of each possible value they will be normally distributed (the bell curve, with most values being near the peak, since all values are equally probable) with any particular value being located at random on the curve -- there is no correlation between a given value and its probability of occurrence.
Gaussian noise is distributed so that the values themselves follow the bell curve with the extreme values at the tails of the curve and the median values in the middle (on average of course, any given trial can give weird results).
So, to check your distribution you could write some values to a table and then bin them according to value to see if the points around zero occur a lot more often than the points near one and minus one.
I tried your patch with [noise~] and [gaussianoise] writing to tables and visually compared the results and it's true you can't hear the difference but you can definitely see it.
(gaussianoise has occasional values that exceed [-1 ... 1], which I suppose is normal...white noise is always on [-1...1])
With white noise there should be no preferred values but gaussian noise will have a lot more hovering around zero.
Martin
Cheers all,
Andy
PD-list@iem.at mailing list UNSUBSCRIBE and account-management -> http://lists.puredata.info/listinfo/pd-list