Stems help when the mix is finished and approved and one element needs a move a stereo file cannot give without dragging everything else along. They do not help when the complaint is balance or masking, because that is a mix revision with a different invoice on it. The test before agreeing to stems: name the move you want, in decibels, on one element. If you can name it, stems are the right tool. If the answer is that the mix does not feel right yet, the file that needs to travel is the mix, back to the person who made it.
What is a stem, and what is it not?
A stem is a grouped collection of sources mixed together by one person and handled downstream as one unit (Stem (audio))). Drums as one file, bass as one file, everything melodic as a third. A stem is not a multitrack, and a folder of 38 files labelled "stems" is a request for an unpaid remix.
The distinction shows up on the invoice before it shows up in the audio. A multitrack delivery means somebody is going to rebalance the record. A stem delivery means the balance is settled.
What does stem mastering actually buy you?
Resolution on one element without collateral damage anywhere else.
A lead vocal that sits 1 dB low across the whole song is the clearest case. On a stereo file the only route to it is a broadband lift around 2 kHz, which takes the snare, the guitars and the top of the piano with it, and then you are fixing the fix. On stems it is a fader move that takes four seconds and touches nothing else.
Two others come up constantly: de-essing a lead vocal without de-essing cymbals, and high-passing the music stem at 35 Hz while the kick keeps everything.
What will stems not fix?
Anything that is already inside a stem.
If the kick and the bass cancel around 60 Hz because the kick-in microphone went up with reversed polarity, that cancellation is printed into the drum file and the bass file, and no combination of level and equalisation recovers what was subtracted at the tracking stage. Masking works the same way: a vocal buried by a rhythm guitar in the same band is only movable if the two sit in different stems, and on most pop mixes they do not. A drum stem printed through a bus compressor taking 6 dB of gain reduction has already lost its transients, and nothing downstream puts back a shape that was never written to the file.
Do the stems have to sum back to the approved mix?
Yes, and the check takes two minutes. Import the stems, sum them at unity with no processing, and invert the approved two-track against the sum. The residue should sit below −80 dBFS. Anything audible in that residue means something was not printed: a reverb return on an output nobody included, a sidechain that behaved differently when the source was soloed, a plugin left in a state the mix engineer did not intend. Running that test before sending is a ten-second fix; running it after the master is approved is a reprint of everything.
How do you print stems when the mix bus has a compressor on it?
This is where most stem deliveries break, and the reason is arithmetic. A bus compressor at 2:1 taking 2 dB of gain reduction on a full mix takes almost nothing on the drums alone, because the drums alone are quieter and differently shaped. Print each stem through the same chain and the sum will not match the mix that everyone signed off.
Two ways out, both honest:
1. Bypass the mix bus chain, print the stems clean, and hand the mastering engineer the chain settings along with the approved two-track. The stems null against a pre-bus sum rather than the final mix, which is stated in the notes so nobody is surprised. 2. Key the bus compressor from the full mix. Feed the detector a copy of the whole two-track while the audio passing through is a single stem, so every pass receives an identical gain curve. The stems then sum to the approved mix exactly, and the null test proves it.
Either is defensible as long as the notes say which one happened. Printing stems through an unlocked bus chain and calling them the mix is the failure case: the master gets built on a balance nobody approved.
How many stems, and how do you name them?
Four to six files, numbered: `01_DRUMS`, `02_BASS`, `03_MUSIC`, `04_LEAD_VOX`, `05_BVS`. Same start point, same length to the sample, no fades, no edits. 24-bit at the session sample rate, peaks on the sum around −6 dBFS, mix bus limiter off.
One text file in the folder saying what is inside each stem. A mastering engineer who does not know whether the tambourine sits in the drum stem or the music stem makes the wrong move on the first pass (what to put in a mix prep folder).
What does stem mastering cost, in money and in revisions?
The Recording Academy put a cross-genre committee on exactly this question through its Producers and Engineers Wing, and what came back is about paperwork rather than plugins. Contracts should state who is paid for creating the stems, who gets credited for them, and what they may be used for, with the cost in the mix budget from the start rather than appearing after the fact (P&E Wing technical guidelines, Delivery Recommendations, revised September 2018). The committee also flagged how easily stems get misused once they exist, which is the real reason to write down the intended use.
The revision cost rarely gets quoted. A stereo recall is one file. A stem recall is a reprint of every stem, a fresh null test and a rebuilt session, and if the mix changed at all the old stem set is worthless (Abbey Road).
When should you just send the two-track?
When the deadline is short, when the mix engineer and the mastering engineer have worked together before, when the label wants the mix locked for credit or contractual reasons, and on anything recorded live where the balance is the performance.
Time counts too. A stereo master of a three-minute single is under an hour. The same record from six stems is closer to two hours before the first revision arrives, because the session has to be built, nulled and documented first. On an album that multiplies by the track count, with the sequencing work from album mastering against single mastering still on top.
Loudness is not a reason either way. The target comes from where the record is going and platform normalisation does what it does regardless (how platform normalisation changes a master). Stems widen the choice of where to find 1 dB, which occasionally means the limiter works less hard for the same perceived level.
I hold mixing and mastering credits on the same records, including "Stefania" for Kalush Orchestra, and the view from both chairs is that an argument about stems is almost always an argument about whether the mix is finished. Those credits with their sources are at /credits.
Frequently asked
How many stems should I send? Four to six. Drums, bass, everything melodic, lead vocal, backing vocals, plus an effects stem only if the record depends on printed delay throws. Past eight files the session becomes a rebalance, which is a different job at a different price.
Do the stems have to sum exactly to the approved mix? Yes, and it is worth proving rather than assuming. Sum the stems at unity, invert the approved two-track against them, and look at the residue. Below −80 dBFS means the print is clean. Anything audible means something was left out.
Is a stem master louder or better than a stereo master? Neither. The ceiling, the loudness target and the delivery files are identical. What changes is how small a correction the mastering engineer can make without moving the rest of the record with it.
If you are deciding between the two
Send the approved two-track first, with the stems in the same folder and one line naming the move you want in decibels. If the stereo file gets there alone, the stems stay unopened and the invoice stays smaller. Background and contact at /about.