Listen to your vocal track, make a mental note of the louder words (the ones you want to compress), but also the quiet words (the ones you don't want to compress).
Now start with a reasonable ratio 2:1, 3:1, 4:1, any will do. Start lowering the threshold until your seeing gain reduction on the louder phrases. The quieter parts should be untouched.
Now critically listen to the vocal, are the transients getting squashed? Does the vocal sound too restricted?
Try increasing the attack. (You may need to lower the threshold a touch more if you increase the attack)
Does the vocals move in the mix? it should be locked in place by the compressor. If so, try increasing the release until the level sounds consistent.
Now, can you notice the compression on the louder words? Try lowering the ratio slightly.
Once your happy with the sound, add some make up gain to get back to the original level.
If after all this, if you feel the vocal needs compressing more, add a second compressor or automate.