My approach for spoken word is simple:
Always have some amount of compression happening at all times that is between 3db and up to as much as 10-12db.
The attack should be as fast as possible and the release set to be nearly fully released between words. The release time takes some listening and looking to nail down. In general the cadence will live right around 180-200ms, but it really depends on cadence.
I like having the compression always engaged so that the effect of pumping is reduced and limited. Try and keep the amount of total compression to no more than 10-12db. Keeping it so that any given word has at least 3db of compression will help the compressor act more as a volume leveler and not seem overdone or clamping things down too much. As with anything in audio, it depends. This is not a set and forget setting, you will have to monitor and adjust it regularly. That is my approach, others have different ways. There is no one size fits all, you will have to try them all so to speak to find what works best for you.