All Amplify knows is that somewhere in your hour long presentation, one tiny bit of blue wave was higher than all the others and it latches on that as the reference for the whole show. If that one little wave happened to be noise, then it will push the desired voice into the mud to accommodate that wav. It has no idea what "noise" is.
I discovered this or something similar yesterday before I had a chance to look back at the forum.
I have been experimenting with removing the clicks, coughs, cellphones & other peak noise & that seems to allow the amplify step to work better & the resulting audio is better-sounding.
I've also found that in a lecture type track, if the speaker coughs or a window shutter falls down or chair squeaks, usually the talking stops for an instant. So I can just cut and past quiet moments from nearby over the bad noise & this blip is unnoticeable. And the peak goes away.
Then I realized it's ALWAYS noise in the context I'm working, so I've been experimenting with the Nyquist limiter. This seems to work ok with the whole track - I can suppress peaks without audible damage to the content. (Sometimes the peaks are laughter or people shouting &c & not appropriate to wipe out.) Then I apply normalization (some of the tracks are noticeably off) & this seems to be able to do what amplify goal does; at least the results look and sound the same to me.
At this point I can do an assessment & see if the rest of the toolkit needs to be used. Some are good enough to stop, but most still have hum and some other noise, needing more work.
Thanks for your help!