When you are producing or remixing, the thought turns up sooner or later: "I want that song’s drums as the foundation of my track", or "what if I took that vocal phrase somewhere completely different?"
That used to be difficult unless the multitracks happened to be officially available. Lately, though, AI-based stem separation can split a song into its parts, and since it arrived there is a lot more you can do.
So this time: how stem separation fits into sampling and remix work — the basic idea, and some practical ways to actually use it in a DAW.
Three things to do with it
Three concrete ideas for using stem separation in production.
1. Take just the vocal and build a remix around it
The first is extracting only the vocal and remixing around it. Pull the vocal out of a song you like, load it into whatever DAW you use — Logic, Ableton, FL Studio, Cubase — and rebuild the chords and the rhythm section from scratch. You can get to something completely unlike the original fairly easily: a club-facing dance remix, or something much more relaxed and lo-fi.
2. Sample the rhythm section into your own material
The second is sampling the drums and bass into original material. Get a clean drum track out and you can chop it up into a kit with its own attack and its own room sound — that particular record’s atmosphere, in your sampler.
Pulling out the bass on its own to study the phrasing and the tone, or to take inspiration from it and write something new, works very well too.
3. Analyse the melodies and the chord movement
The third is analysis. Besides the obvious parts, keys, synths and guitars often arrive together in an "other" category.
Listen closely and you can pick out things the vocal and the rhythm were covering: little obbligati answering the other parts, and the particular voicings the chords are stacked in. For taking apart how a professional arrangement is actually put together, by ear, it is enormously useful.
Three tricks for making them sit in your track
Now, when you drop those separated tracks into a DAW, on their own you may hear the characteristic hiss that separation leaves behind, or a slight wobble up in the highs. Three small things help them settle into your own material.
Trick 1: blur the edges with reverb and delay
The first is making good use of space — reverb and delay. A thin reverb or delay on an extracted vocal or phrase blurs the fine noise separation leaves behind, and the unnatural way a phrase sometimes stops dead. It starts to sound as though the part was always sitting in that space.
Trick 2: cut the bands you do not need
The second is an EQ, cutting away bands you have no use for. On an extracted vocal, a low cut clears out whatever kick drum from the original backing came along with it, and the whole mix gets tidier. Just giving each part its own useful range pulls everything together.
Trick 3: let your own tracks be the loud ones
The third one matters more than it sounds: keep your own tracks properly loud. Do not push the sampled material too far forward — give the drums and bass you programmed yourself the level they deserve, and the separation noise stops being noticeable in relative terms.
It helps to think of it as: your track is the lead, and the sampled material is colour and spice.
In short
With stem separation, ideas that used to stay ideas — because getting a clean part out was too hard — can be turned into something almost immediately.
If you want to widen the range of what you arrange, or work up new ideas with an existing phrase as a reference, or just have a go at remixing without much ceremony, a free browser-based stem separator is worth an afternoon.
The stem separation in this article runs free in your browser — no sign-up, and your audio never leaves your device.