There are numbers of way to do it, I think the best for you would be to have 2 vocal tracks getting the same mic input, set up one that has effects for singing and another for talking perhaps with a compressor and an EQ just for the speech. Then mute the speech mic track all the way and whenever you need to talk, solo the muted track (note that you can solo muted tracks.). You can also write automation to enable/disable them during a song.
Another simple solution is to ALT+Click on the insert indicator icon on the project window, it will bypass all the insertions. You’ll need another click for sends and also you can of course forget to turn them back on when you start next song, so I think the first one should be easier and safer for liveuse. 1st idea won’t play music even if you accidentally pressed play when the talking mic is on.