This is a reply to private mail from Eric. He asked to post it here
such that we all can see it. Disclaimer: if my language sounds rude, it
is such because my english is not that good.
The power function used by various gain functions described in the specs
has something wrong because the square root should be taken from
non-negative values only. Also, in "pcompress" the p is assumed to be
between values -1 and 1 even the square root gives only non-negative
values.
My best bet for the new power function is that the signal value should be
powered by 2 before the sum. I don't know if there is some constant
gain in front of the whole expression. I forgot my notes home containing
both an integral expression and an FFT based expression for the power.
The latter is sqrt(a_0 + a_1^2 + ... + a_n^2) where a_0 is FFT component
of the DC and a_i's are other FFT components.
I wrote about the commercial compressors:
Some commercial audio programs such as Sound Forge and its plug-in Ultra
maximizer (or such) has quite different looking compressors and limitters.
Would you please tell how close "compress" and "pcompress" comes to those
commercial ones? Are they even meant to be close to commercial ones? I.e.
what makes "pcompress" to be what it is now -- is it such just for
historical reasons?
Eric replied and basically said the ones in Saol are strange.
>Do you have expertise in designing parametric compressors? If so
>(I don't!), perhaps you could make a concrete suggestion as to
No. But I have already examined various compressors in commercial programs.
I will look at them and all material related to compressors as soon as
possible -- I will put the material available. Perhaps we find more people
to look at them; I remember already one in Israel who also reads Csound
mailing list -- he has written some paper (which I don't have) on the topic.
I also wrote about the noiseshaping ditherer. I have material for such too.
I feel the noiseshaping ditherer (or plain ditherer) is important to get to
the core because if it is in a library, users may not use it if they don't
understand its meaning. In my opinion, the noiseshaping ditherer should
be applied always (without asking from the user) when the signal is put
out with 16 bits only. In implementations of the Saol, there could be
a HiFi mode using the ns. ditherer and some low quality mode not using any
ditherer at all.
I also requested if it is possible to modify the example implementation
to read audio from stdin and write to stdout such that we may use the
program "as is" for making effects in upcoming Audiotechque, a free wave
editor program.
Juhana
This archive was generated by hypermail 2b29 : Wed May 10 2000 - 12:14:09 EDT