On Sun, Sep 26, 2010 at 08:52:30AM +0200, Philipp Überbacher wrote:
> What's an ad-hoc algorithm? Why are general purpose audio
When you compress or limit a voice or instrument, the
intention is to either *modify* its sound, or to reduce
its dynamics so it becomes easier to integrate into the
mix. The exact technical level at the output of the
compressor is not really an issue in that case. The
operation will be followed by EQ, effects and other
tracks will be added, so the actual level doesn't
matter that much. A compressor set for 'infinite'
ratio (or a limiter) will output signals well above
its threshold - it has finite attack time, and even if
set for 'peak' detect it will still do the wrong thing.
Peak limiting as the last step in mastering (to ensure
that the signal stays within the [-1..+1] range uncon-
ditionally is quite a different matter. You operate
on a mix, the artistic decisions have been taken and
you don't want to really modify the sound - just raise
its level as much as possible while staying within the
purely technical limits imposed by distribution in a
digital format. And the limits in this case are *hard*
- no single sample should go outside the allowed range.
Single isolated samples exceeding the limits can in most
cases just be clipped. Short overloads (< 1ms or so) can
be handled by looking ahead and operating only on the high
frequencies. Longer ones can be handled by more conventional
limiting algorithms. There are good reasons to make attack
and release times dependent on spectral features, amount of
overload, recent history etc. So any limiter trying to do
this job would need a mix of several special algorithms.
You can go some way by making the operation multi-band.
But a mult-band limiter put together from separate plugins
can't really do this. First, none of the components has
any idea of the final (summed) output. But that is the
signal we want to control quite accurately - the separate
outputs really don't matter here. Second, the components
don't know what the others are doing - they can't coordinate
in any way. There is nothing that prevents e.g. the LF limiter
to reduce gain by 10 dB while the rest of the frequency range
is left untouched. What you want in such cases is that the MF
and HF limiters at least apply part of that reduction as well.
Doing all this right requires a limiter designed for the job.
There are three of them, and Alleline.
Linux-audio-user mailing list