[music-dsp] Adaptive (auto) release

[Citizen Chunk]

hi all. i've been making great progress on my compressor project 
(thanks in part to this group), and i'm trying to tackle adaptive 
release control. here's the deal:

i'm monitoring peak and RMS (via a simulated RMS equation) on a per 
sample basis. i plan to have the envelope detector monitor the peak 
value, but use the ratio of peak:RMS to control the release timing. 
short transients (larger peak:RMS ratios) are tracked using a "fast 
release" time. longer sustained waves (smaller peak:RMS ratios) are 
tracked using the "slow release" time. the release could simply 
cross-fade from fast to slow based on the current ratio.

the problem (unless i'm looking at this wrong) is that, while the RMS 
remains fairly constant, the peak will fluctuate between being above 
and below the RMS, even though the signal is rectified. isn't this the 
case? as such, that would throw off my equation. here's what i was 
thinking:

relFast = fast release time
relSlow = slow release time
peak = fabs (input)
RMS = well, you know what that is...

release = relFast + ( (relSlow - relFast) * (RMS/Peak) )

assuming that the peak level is always above the RMS (and never = 0, so 
as to avoid divide-by-zero), this would be a nice little equation to 
cross-fade between the fast and slow release timings, 
program-dependent, of course. when the ratio approaches zero, the 
release approaches relFast. as the RMS approaches Peak (and the ratio 
approaches 1), the release approaches the relSlow. the problems are 
those pesky conditions i just mentioned.

any thoughts of how to remedy this? or am i looking at the whole 
adaptive release control issue the wrong way?

btw, i know that this is one of the things that separates competing 
software, so i'm not asking for people to give me their top-secret 
algorithms. just a kick in the right direction.