Stem mastering vs stereo mastering: when splitting the mix is worth it

Stereo mastering is the default: you finish a mix, bounce one stereo file, and the mastering engineer (or you) works on that file and nothing else. Stem mastering sits between mixing and mastering. Instead of one file, you send a small set of submixes, a handful, that add up to the whole song. The engineer can then turn the vocal down half a decibel or put a little compression on the drums alone, moves that are impossible on a stereo file without affecting everything else.

This guide explains the difference, when the extra step is worth paying for or doing yourself, and how to print stems that do not cause more problems than they solve. If you are preparing files for an automated service rather than a person, our guide to prepping stems for AI mixing and mastering covers what those services ask for. If you are mastering a stereo file yourself, start with mastering at home.

What changes, and what does not

In stereo mastering, every process touches everything. A cut at 250 Hz to clean up a muddy guitar also thins the vocal and the snare. A limiter reacting to a loud kick also ducks the pads. The engineer's job is to find moves that help the whole picture more than they hurt any part of it.

With stems, those trade-offs loosen. A typical split looks like this:

• Drums (kick may be its own stem in bass-heavy genres)
• Bass
• Music (keys, guitars, synths, pads)
• Lead vocal
• Backing vocals and effects returns, sometimes split again

What does not change is the output. The deliverable is still one stereo master, measured the same way. Spotify, for example, normalises playback to about -14 LUFS integrated and recommends keeping true peak below -1 dBTP (below -2 dBTP for masters louder than -14 LUFS), using the loudness method in ITU-R BS.1770 (Spotify for Artists). Apple's Apple Digital Masters programme asks for high-resolution masters made by engineers. As we read it, the programme is about the final master file, and nothing in these delivery specs turns on whether stems were involved.

When stem mastering actually helps

There are a few honest reasons to work from stems:

1. You cannot fix the mix any more. The session is lost, the mix engineer is gone, or the project was made on hardware that cannot be recalled. Stems are the last point where balance can still be changed.
2. One element is fighting the master bus. A bass that makes the limiter pump, or a vocal that jumps forward every time the track gets brighter. Processing that stem alone solves it cleanly.
3. You mixed in a room you do not trust. If your low end was guessed on headphones, giving the engineer the bass and kick separately lets them correct it without re-EQing the whole song. Our guide to mixing on headphones explains why this happens.
4. You need alternate versions anyway. Instrumentals, TV mixes and a cappellas for sync are easier to deliver consistently when the master was built from the same stems.

And a few bad reasons: hoping the engineer will "mix it for you" at mastering prices, or sending forty stems because you could not decide. In our view, past a handful of stems the job stops being mastering and becomes mixing, and it is fair for an engineer to treat it that way.

How to print stems that sum back to your mix

The single test of good stems is this: put them all in a new session at unity gain and the result should match your stereo mix. If it does not, the engineer starts from something you never approved. Step by step:

1. Freeze the arrangement. No more edits. Note the tempo, sample rate and bit depth.
2. Decide the groups. Route tracks to buses that match the stems you will send. Our guide to bus processing and groups covers the routing.
3. Take the master bus processing off, or print it into each stem knowingly. A bus compressor or limiter on the master reacts to the sum. Solo a stem through it and the compressor behaves differently, so the stems will not add up. The cleanest option is to bypass master bus dynamics, export the stems, and tell the engineer what was there.
4. Keep effects returns honest. If the vocal reverb lives on a shared return, either print it as its own stem or bounce the vocal with its sends. Do not lose it between files.
5. Export every stem from the same start point to the same end point. Every file should begin at bar 1 (or 0:00) and run to the end of the longest tail, even if it is silent at the start. In Ableton Live 12, the Export Audio/Video dialog can render all individual tracks in one pass, which keeps lengths aligned (Ableton manual). Other DAWs have an equivalent batch or stem export.
6. Use 24-bit or 32-bit float WAV at the session sample rate, with no dither and no normalising.
7. Null-test. Import the stems, sum them, flip the polarity of your stereo mix against them. Near silence means you are done. Audible music means something on the master bus, a sidechain or a shared return did not make it into the stems.
8. Send the stereo mix too, labelled as the reference. The engineer needs to know what you thought the song sounded like.

Stereo mastering is still the better default

For most releases, stereo mastering is quicker, cheaper and safer. It forces the mix to be finished before mastering starts, and the engineer's fresh ears work on the song as one thing, which is closer to how a listener hears it. If you find yourself wanting stems for every release, that is useful information: the problem is probably upstream, in the room, the monitoring or the habit of calling mixes done too early.

What we think

Stem mastering gets sold as a premium tier, as if more control always means a better record. We think that framing gets it backwards for most independent artists. The value of a mastering engineer is a second, trained opinion on a finished piece of work. Handing over stems quietly moves the line between "my decisions" and "their decisions", and the balance of your own song, the most personal thing about it, drifts into someone else's hands.

So our rule is simple. Use stems when you are rescuing something, or when you have a specific, named problem you cannot solve in the mix. Otherwise, fix the mix and send one file. When you do send stems, keep them few and treat the null test as non-negotiable, so that whatever comes back is still your record, only clearer. The artists who come out of the stem route happiest are the ones who used it to protect their mix, not to finish it.