Powair comes very close to being my favourite compressor/leveller for dialogue. The only problem I find is pumping sometimes going from quiet to loud passages and the issue with the noisefloor which other have commented on. I had a thought that if an AI algorithm that could recognise certain sounds as dialogue was incorporated into Powair, or any other compressor for that matter it could work only on the Dialogue parts and leave the rest of the sound untouched, especially when using look ahead on already recorded material. I have no idea whether this is actually feasible, but it seems like it could be promising if someone was to develop it.