Why is 15 a reasonable mask-level value in the example Help->Browser...->PureData/->3.audio.examples/->I04.noisegate.pd? Mask-level multiplies the frequency-indexed average noise levels contained in the mask table before they are used as gate thresholds. I'd expect that 1.5 would suffice and that 15 would exaggerate the gain reduction so much that it would leave unintended artifacts, yet that's not what I hear. For values between 0 and 4 I hear mp3-like chiming which goes away above 6. Miller Puckette discusses the algorithm in Theory & Technique of Electronic Music ch 9.4.1, but there's no mention of scaling the threshold function f[k]. The only thing I can think of is that maybe the distribution of noise powers for each frequency over time is so broad that you need a factor > 6 to include most of them?
Update: if I bypass the q8_sqrt~ things behave as I expect them to. I don't understand why he's taking the sqrt anyway--there's no mention of it in the gain equation g[m,k].