Try and match the RMS (Average) levels and don't worry about the peaks as long as they are below a sensible limit.

If you are using PT go to the gain plug-in, and you can find the RMS level there.

What I usually do is find a region that I consider to be dynamically perfect for the section I am working on, i.e. just the right balance between RMS and peak. You don't need to be too precise with the RMS match, within 2 or 3 dBs is usually fine.

I will then try and replicate the dynamic range on the problem regions using a compressor and make-up gain. So if the dialogue involves whispering and shouting, then the difference between the RMS and peak will obviously be greater, or if the dialogue is supposed to be monotonous, then the difference between RMS and peak will be much less.

The way i do it is to ride the faders to balance things before going into eq, de-esser and compressor. Then i put a Brickwall limiter on the submaster and master. It's timeconsuming, but it's worth it.

You could try the Waves Vocal Rider plug if time is more important than quality in this case.

have a think about the characters, sometimes they're supposed to vary like that.

up the gain on the quieter voice before compression, lower the gain on the louder before compression. find a place to insert an EQ on each, before compression. then have a think about the microphones used. on the louder voice you may need just to shave the peaks off, whereas on the quieter you would be less subtle. watch out for noise, and don't look at the meters too much.

You could try an automatic gain rider, e.g. http://www.meldaproduction.com/mautovolume/ and adjust the settings for the two tracks by ear.

You could also try an upward/parallel compression approach. Dupe the "too quiet" audio onto a second track, add a compressor with a low threshold and high ratio (so that it's really compressing the hell out of the dialog), fast attack, slow release, and slap on some heavy make up gain. What you're left with is the a track full of the lower end of the dynamic range. Route the original and the parallel compression to an aux channel and mix them to taste (you'll probably only use a little of the parrallel track). Make sure you don't have any phase/timing issues being introduced by the plug-in, adjust the timing of the parallel track if you do. This should help add some body to a weak track. If any additional compression is needed, add it to the aux track the two copies are being mixed to.

It has already been mentioned, but its an amazing (Pricey) tool. If you go through this often, then this will make your life so simple.

http://www.waves.com/content.aspx?id=9945