Stems, explained: what splitting a song can and cannot do

Stems is the first specialist word Orbit Waves teaches. A stem is one part of the song. Most splitters give you four: voice, drums, bass, and the rest.

Why split at all

A finished stereo file is baked. If the vocal is too quiet, an EQ can only push a frequency range, and the guitar in that range comes up too. With a separate voice stem, the Voice slider moves the voice and leaves the rest alone.

Two kinds of split

Neural separators are trained models that rebuild each instrument. They give the cleanest parts and need a lot of computing power, often a GPU.

Filter-based splits divide the song by frequency and stereo position: lows to bass, centered mids to voice, side highs to drums. They are fast, run on a normal laptop, and give you parts that are good for balancing. They are not clean isolations: some of the guitar will live in the voice stem.

Orbit Waves has both. Split into parts is the fast filter: it runs on your computer in seconds and the four parts always add back up to the original song. Cleaner split on our server runs Demucs, an open neural model, and takes about a minute. Free accounts get 2 server splits a month, the monthly plan 150.

When the split is enough

If you need the voice to sit higher in your own song, the fast split and one slider do it. If you want parts you could almost play on their own, use the server split. If your AI tool lets you download stems directly, drop the zip into Orbit Waves instead: each file becomes its own part.

• Turning the vocal up or down a few dB
• Tucking a boomy bass
• Muting the drums for a moment, or checking what each part is doing