Errrr, uuhhhhh, yeah, what you said.
;-)
Makes sense. I will play with speech signals and the function using
capture and matlab to see what
makes sense but my guess is you are pretty close to right. I was indeed
thinking Gaussian noise
filling the envelope curve. You caught me again trying to cheat.
Bob
Frank Brickle wrote:
n4hy wrote:
...I am going to figure out the best regression line at the place of
maximum slope on this compression curve. That slope will tell you
how much gain is being applied at that agnitude. If it is 10 dB, I
am going to call it "10 dB of compression"...
The only problem I have with this is that it's sensible for gaussian
input. In fact, you know very well it's correct for gaussian input :-)
However, since it's applied almost exclusively to speech input, the
signal isn't gaussian in the time intervals where it matters most. I
kind of suspect a better measure is the integral in some window around
the average input RMS.
So I'd speculate a better empirical guess would be based on the
convolution of the waveshaping function and two convolved gaussians.
Too hairy for me. Why not the gain applied at the tangent where the
slope = 1/2?
73
Frank
AB2KT
