Labels: stems before you duck the vocal

Split, then duck.

A label mix that still has the singer in the bed is not a duck setting. It is the wrong file. Ducking a full bounce under a voiceover asks a sidechain to solve an arrangement problem. Pull the bed down and the hook reappears every time the VO breathes. Music stems before you duck the voice is the order of operations. This page is the label version: the vocal is the catalog. Get parts first.

A bounce is already a record

A stereo song is mixed to be complete. A VO, a trailer line, or an interview bed needs a hole. Ducking changes level over time; it does not remove a part. If the part that collides with speech is still in the clip, the collision follows the duck: quieter during lines, present in every gap.

Two generation choices prevent that:

1. Never write the lead. Prompt or flag the generate as instrumental. No vocal, no vocal to duck around.
2. Write the lead, then mute it. Split stems. Attach the instrumental as the bed. Keep the vocal stem for the cut that is actually the record.

Do this at generate time, not at 11pm in the NLE. A label that "will stem later" previews the bounce, locks picture to it, and then never replaces the file.

Two splitters, two objects

Split a song into stems is separate_music_vocals. It splits a track you generated in-app into vocal and instrumental stems. You need taskId and audioId from a prior generate_music or extend_music result. It is not for an arbitrary uploaded file. Billed per split, on top of the original track's generation cost. Confirm with estimate_cost.

For a mixed recording that did not come from that generate — an interview, a bounce someone emailed — isolate vocals from a track is isolate_audio on any audio_url. You get a clean vocal with the bed removed. That is the general-purpose separator. It is not a 4-stem session. Separation quality depends on the source mix.

Uploaded commercial tracks are a licence problem first and an isolate_audio problem second. Do not stem a record you cannot clear. Ask the label for an instrumental or stems. If they do not have them, you are mixing a bounce: keep it quieter than you want, accept the gaps will bloom the hook, and do not promise an M&E.

ElevenLabs Music v2 has its own 2 / 4 / 6-stem split when that is the generator. Two stems solve "no singer."

Duck the bed, not the singer

Decide whether music is the show. If the clip is a chorus-led Short, you want the bounce. Stop. If the clip is a talking product, a founder line, a tutorial, music is furniture. Furniture does not sing.

Generate furniture without a lead, or split the approved cue. Attach the instrumental, not the bounce. Mix mode, not replace. Static music_volume well under the voice. Duck only after the lead is gone — fast attack, slow release. You are lifting furniture in the holes, not hiding a singer. Then hit a loudness target. Mastering the stereo bounce first bakes the fight into a file with no headroom to carve.

Do not duck, then stem, then duck again. Do not treat music_volume: 20 as stems. That is a static fader on whatever file you attached. If you attached the song, you ducked the song. Do not generate a new track to "fix the mix" when the mix is a lead-vs-VO problem. You will get a new lead.

A leaked ghost vocal on an instrumental stem is a reason to listen solo before you lock. If the ghost reads as a second speaker, regenerate instrumental.

FAQ

Can I duck first and stem only if it sounds bad?

You can. You will do the attach twice. Preview a stem. Do not preview a bounce you already know you will throw away.

Can I split any master the label sent?

separate_music_vocals only works on in-app generates, referenced by taskId and audioId. For an arbitrary upload, isolate_audio. For a commercial master, clearance first. If they have no instrumental, you are mixing a bounce.

Is an instrumental generate always better than a stem split?

When you know there is VO, yes — nothing to leak. When a vocal version is already the approved cue, split it. Do not regenerate a new instrumental that no longer matches the approved cue's harmony just to avoid a split.

Does this replace carving 250–500Hz on the bed?

No. Stems remove the part. Carve removes the band. You still carve an instrumental. You just stop asking a duck to mute a singer.