Stems are stereo (or mono) subgroupings of instruments or vocals. For example, the most common stem mixes have a stereo instrumental stem and a stereo vocal stem. When these two stems are set to unity gain (no volume boost or cut) and played through a common stereo output, you should have the complete stereo mix as the mixing engineer/producer intended it. But now if the mastering engineer needs to brighten up the vocals for instance, or thinks the guitars have too much midrange, then he can work with one stem without affecting the other. You might also have a case where the bass has been separated out to a mono stem of its own, so the mastering engineer can compress or eq it independently of everything else.
Stem mixing is really very simple. You don't do anything different until you're ready to bounce your mix. Mix as you normally would, get all of the instruments and vocals balanced, eq'd, panned and f/x'd until you're happy with the mix. Then mute all of the vocals and bounce your mix. You now have the instrumental stem. Then, unmute the vocals, mute everything else, and bounce again. You now have the vocal stem. If you load the two stems back in, set them for exactly the same start time, and play them both through the stereo output, voila! You are hearing your stereo mix just as you wanted. If you have done things correctly, listen to the instrumental stem and you will hear that all of your automated fader moves, f/x etc. have been printed with that stem. You should hear no vocals, vocal reverb, echo or anything else related to the vocals. Likewise if you listen to the vocal stem there should be no hint of instruments or instrument f/x.
So you see it would be the exact same procedure if you wanted to separate out bass, background vocals, drums, whatever. You just mute everything else in the mix except those elements and print. But the mastering engineer would prefer you keep it as simple as possible, the ideal being an instrumental stem and a vocal stem as in the example above. Don't distract the mastering engineer by delivering too many stems in one song so that he is suddenly having to listen like a mix engineer and not a mastering engineer (two VERY different ways of listening, btw).
If you are mixing through an analog console with analog f/x, you need to be very conscious of processor noise. When you are printing your vocal stem, you should mute all f/x processors that are dedicated to instruments only, or if you have dedicated returns for instrument f/x then those should be muted. Otherwise you will be printing the noise from your instrument f/x processors onto the vocal stem, and then you will get the noise again when you print the instrumental stem. Although it may seem like a minor detail, you should do the same when you print the instrumental stem regarding the vocal f/x returns. Even on loud tracks, every little bit of noise you can eliminate will make a difference in the final product.
In the hip-hop world, delivering stems for a single can make your life much easier. All you need to deliver is the instrumental and vocal stem for each song. Now the mastering engineer can add the two together for the club version, mute the bad words on the vocal stem for the radio version, instrumental stem only for instrumental version, vocal stem only for acapella version. Too easy. (Or save a little money and do a clean vocal stem yourself if there are a lot of words that need to be muted).
Make sure you talk to your mastering engineer before you deliver stems. Not all mastering rooms are as technically hip and savvy as DES Mastering (shameless plug), so you shouldn't assume that they can work with stems.
