Two-stem vs four-stem vocal separation: what Spleeter modes mean for your download
Search queries about two stem vs four stem vocal separation usually mean: "Will four stems give me a cleaner instrumental?" Sometimes. Often you only need two. The answer depends on what you are building, not what sounds more premium in a feature list.
Two-stem mode: vocals + accompaniment
Spleeter's 2-stem model splits the mix into vocal and non-vocal energy. That is what SongRemoveVocals delivers — instrumental and vocal downloads for karaoke, video beds, and remix starting points. Faster inference, lower GPU cost, predictable output for the majority of "remove the singer" jobs.
Four-stem mode: vocals, drums, bass, other
4-stem separation adds drum and bass channels plus an "other" bucket for guitars, keys, and residual bleed. Useful when you want to mute drums for a stripped arrangement, pull an isolated bass line, or rebuild a mix piece by piece. Not automatically cleaner on the vocal/instrumental divide — the vocal stem can still carry snare bleed on dense rock.
Why we default to two stems
Our product is vocal removal for people who need a backing track or acapella-ish stem without operating a stem laboratory. Two-stem Spleeter on Replicate balances quality, latency, and the economics behind 10 free minutes per day and minute packs from $4.99. Four-stem runs cost more compute per minute; that cost lands in pricing or free-tier limits.
Model history in Demucs, Spleeter, and cloud separators. Local tools like UVR expose multiple modes if you want to experiment offline.
Pick the mode by job
- Karaoke / video instrumental → two-stem instrumental is the deliverable.
- Mashup acapella → two-stem vocal stem; expect cleanup per acapella vs vocal stem.
- Drumless practice track → four-stem (elsewhere) or mute drums after 4-stem export.
- Re-release quality → neither mode; license official stems.
Upload limits either way
MP3, WAV, M4A, FLAC — 15 MB, 5 minutes max per file. Export WAV or MP3 after processing. Refer a friend for 10 bonus minutes when A/B testing the same song across tools.
Run your material through SongRemoveVocals and compare against other online options on the chorus, not the intro.
Related reading
- Remove vocals from iTunes and Amazon purchased songs: format, DRM, and expectations
- Minute packs vs subscriptions: pricing for people who separate occasionally
- Why vocals bleed through (even with AI separation)
- Online vocal removers compared: Moises, LALAL.AI, Vocal Remover, and SongRemoveVocals
Disclaimer: Stem count describes output channels, not separation perfection. Four stems do not guarantee cleaner vocals on reverbed or dense mixes.