Kev999
Certainly ducking is the way to do it, but it's not always necessary to duck everything. Some instruments don't need to be ducked if their freqency range does not overlap with the dialog. I would arrange the routing so that the instruments that need to be ducked pass through a single bus and then apply ducking to that bus, with fast attack and release. If some instruments are particularly distracting, then I would duck them separately, and less subtly, with slightly longer release time.

It is very rare to have access to the music in stem form in order to process only certain parts of the music.  It does work a little more like that in major films but rarely in a TV series.  Maybe in some of the bigger series but often not.  The fact is the there should be nothing distracting to start with under any of the dialogue if the composer has done their job well.  You often not only get an edit of the scene but the dialogue is usually there to alert you where it comes in goes out.
 
From experience ducking seems to be more a radio thing and in many of the docos I was involved with they never ducked the music either.  They automated it.  Sometimes a music cue can lower its level quite slowly but then comeback up a little faster.  If you are ducking you have to really set those attack and release parameters well.  Often the attack is not slow enough either to create the effect of the music easing down as opposed to jumping down.
 
Kev's approach is good though if you are creating the music in your DAW and you have the video playing and even be lucky enough to have the dialogue track playing at the same time.  But instead of ducking elements down in order to stay out of the way of the dialogue why not remove them all together.  Re arrange the music so the melody lines sort of come to rest just prior to dialogue and picking up after it stops.  This is still way better than ducking.  Because the offending sounds are not even there to start with.  Dialogue will often be in large blocks.
 
Also if you are working with production library music often the better libraries offer an underscore version and the tracks will be identical time wise.  Except all the melodic information has been removed.  You can always have both on your time line and cross fade from the melodic tracks to the underscore tracks quite nicely and all seamlessly too.  I have done that before too.
 
Another thing to watch is effects too.  If there are huge effects things going on you may want to keep the music very simple at that point too.

I think Jeff said everything that can be said on this topic. So will do nothing more than recap.
1. The best solution is be able to manipulate the original source material so that you can elegantly remove the lead/solo/'singing' parts from the underlying backing tracks during the dialog sections. 
2. Short of that, or possibly in conjunction with it, you can do selective frequency ducking on the source material so as to allow sonic space for the dialog. This ducking can happen either via an actual frequency dependent ducker or via automation envelopes with a good EQ plugin.
3. A combination of 1 and 2 is probably your best bet.

I thought someone would have mentioned Vocal Rider from Waves as well.
Simplest. It’s on sale right now for 99.00 at the Waves site.

Passive Drift
I thought someone would have mentioned Vocal Rider from Waves as well.
Simplest. It’s on sale right now for 99.00 at the Waves site.

Interesting.
Does anyone know whether Vocal Rider is the same thing as Sidechain / ducking compression, with other name? 
Or better? Or worse?

Vocal Rider is about keeping a constant vocal level, not ducking.

Anderton
Vocal Rider is about keeping a constant vocal level, not ducking.

This would not be the same as compression?

rogeriodec

Anderton
Vocal Rider is about keeping a constant vocal level, not ducking.

This would not be the same as compression?

Not exactly.  It is supposed to be more like riding the volume control to maintain the same level (but doing the work for you and more efficiently).  Compression is a different effect.  
 
I tried it a while ago and that is what I recall.  

Compressors were originally designed to automatically ride the volume to even out level fluctuations
 
It was only later they were employed as an effect in their own right as engineers started to add controls like attack/release etc, providing the opportunity to reshape the dynamic envelope.

I'm a little confused. Unless you have access to have the music written in a way to naturally leave room for narration vox there really isn't any way to practically do what you're asking without lowering the music volume (via whatever technique works). I'd reconsider your premise that the music always needs to be out front in relation to your narration unless you have a wacky plan for one or the other (doing an experimental treatment on the narration vox for example).
 

...wicked
I'm a little confused. Unless you have access to have the music written in a way to naturally leave room for narration vox there really isn't any way to practically do what you're asking without lowering the music volume (via whatever technique works). I'd reconsider your premise that the music always needs to be out front in relation to your narration unless you have a wacky plan for one or the other (doing an experimental treatment on the narration vox for example).
 

The OP is really just asking how to automate the process without having to sit through some long track pushing the volume up and down when each clip of narration start and stops.  
 
The volume changing is the effect he wants, it is the time consuming way of doing it that he is trying to avoid.  Thus why you see responses related to processes that automatically make such changes based on certain variables.