The producer has one stereo WAV.
The original session vanished with an old laptop. The vocalist moved abroad. Nobody can find the instrumental. And a sync request just landed — the music supervisor needs a version without vocals by tomorrow morning.
A few years ago, that was the end of the conversation. Now the producer drops the master into a DAW, hits Separate Stems, and pulls out versions of the vocal, drums, bass and instrumentation in minutes. The same tool works just as happily on a Beyoncé record, an obscure 1970s single, or a song lifted from somebody else's catalog.
That's exactly where convenience turns into a rights question.
Stem separation has moved from specialist restoration software into ordinary production environments — Ableton Live, Logic Pro and Cubase all now pull a mixed file into components you can use to rebalance a master, make an instrumental, study an arrangement, build a live set or start a remix. The software can process the file. It doesn't decide whether you have permission to.
Stem separation is becoming a standard DAW function
Stem separation — sometimes called source separation or demixing — uses a computational model to estimate which sounds inside a finished mix belong to which category. Worth being precise about one thing up front: it does not recover the original multitrack session. It creates new approximations from the information sitting inside the mixed recording.
The tooling has arrived fast. Ableton introduced built-in separation in Live 12.3, running locally and splitting audio into four groups — vocals, drums, bass and "other"; Live 12.4 added processing of selected sections and combining the rest into one track, which makes vocal removal quicker. In July 2026, Ableton also expanded the workflow on Push with high-quality and faster separation modes. Apple's Stem Splitter goes further on categories — vocals, drums, bass, guitar, piano and other instruments — on Macs and iPads with M1 or later silicon. And Cubase 15 includes AI-powered separation aimed at remixing, rebalancing and sampling.
The direction is obvious: separating a record will soon feel as ordinary as time-stretching it. The result still needs to be heard critically — which is the part the hype skips.
An extracted stem is not the original stem
Picture a vocal sitting next to distorted guitars, cymbals and a long stereo reverb. The separation model has to estimate which frequencies and transients belong to the singer and which belong elsewhere. When several sounds share the same space, it can't reopen the session and hand back the clean vocal file — that file doesn't exist as a separate object inside the stereo master.
So the extracted vocal may carry cymbal residue, pieces of the instrumental reverb, missing consonants, unstable highs, phase changes, softened transients, fragments that flicker in and out, or textures the process invented. Some of those flaws only reveal themselves after compression, tuning or heavy processing. A separated instrumental can sound convincing under the vocal and strangely hollow when heard alone, because removing the lead voice also strips frequencies from guitars, synths or backing vocals the model associated with the singer.
None of that makes the technology unusable. It changes the standard. A rehearsal instrumental and an official release aren't held to the same bar — an artifact hidden beneath a new arrangement may be fine, while the same artifact exposed in an archival vocal can become the most noticeable sound on the record. So keep the untouched source file and compare every result against it, and treat separation as a new production stage, not a perfect reversal of the mix.
The Beatles show the best-case use
The most famous stem-separation story started with material the team was authorized to use. Peter Jackson's WingNut Films developed the MAL audio technology while working with Beatles recordings — a system that could demix a mono soundtrack and isolate instruments, vocals and individual speaking voices. That work helped enable the 2022 remix of Revolver and, later, the completion of "Now and Then" from John Lennon's home demo, where the technology let Lennon's voice be separated from the piano on the original cassette.
That was an extraordinary restoration problem — and it involved the artists, estates, producers and rightsholders responsible for the material. The lesson isn't that AI made an old recording fair game for anyone to reconstruct. It's that separation unlocks legitimate creative work when the source, authority and purpose line up.
Independent catalogs hold smaller versions of the same problem all the time: a lost session, a live recording with no isolated vocal, a mono archive, an unfinished demo, a master that needs an instrumental for sync. Those are excellent uses — provided the person commissioning the work actually controls the recording.
Owning the song is not the same as owning the recording
A separated file can contain two different categories of protected material. The composition covers the music and lyrics. The sound recording covers the particular recorded performance and production. The U.S. Copyright Office treats them as separate works — a songwriter may own the composition while a label owns the master, and a producer may possess the WAV without owning either layer.
This matters because separation doesn't replace the original performance. If you extract a vocal from an existing master, the result still contains sound taken from that recording. Calling it an "AI stem" doesn't transform it into newly generated, royalty-free material.
Whoever owns or controls the master may be able to authorize restoration, remixing or extraction on the recording side — but composition approval can still be needed if the project changes the underlying song, creates a derivative arrangement, or folds it into another work. Contracts can narrow that authority further: an artist might own the master but need a featured performer's consent before releasing a remix; a label may control distribution while an approval clause gives the artist a say; a producer's agreement may restrict isolated use of a performance. Open the contract before you open the splitter.
What you can usually do with your own catalog
When you control or have properly licensed the relevant master and composition rights, separation solves real problems.
Recover an instrumental for sync pitching, live use, karaoke-style content or promo video — but check it hard for vocal residue. A supervisor asking for an instrumental expects the lead genuinely gone, not haunting every chorus.
Restore an old recording — isolate speech, reduce crowd noise, rebalance instruments, repair archival material. Keep a record of the source, the software and model, the extracted files and every edit; those decisions matter later when someone asks which version is historically original.
Build a remix from workable material where the original stems are missing — but the remix still needs authorization from whoever controls the necessary recording and composition rights. Permission to "use the song" doesn't automatically include the master, the isolated performances, and commercial distribution of a remix.
Prepare live material — pull the part you'll perform onstage, build backing tracks from your released masters. Review agreements with band members, session players and featured artists if isolated performances will be exposed or reused.
Create alternate versions — a clean edit, a stripped mix, an instrumental, a spatial mix — but give each one a purpose. Separation makes it trivial to generate files nobody needs, and catalog size isn't catalog value.
What you can't assume about somebody else's recording
Buying a download, subscribing to a streaming service, or finding a song inside your DAW does not grant commercial remix rights. And "No copyright infringement intended" solves precisely nothing.
Privately experimenting with a track, studying its arrangement, testing a separation tool — that's one thing. The analysis gets serious the moment extracted material is sampled, distributed, monetized, performed publicly, handed to a client, or synced with video. If the new production carries recognizable audio from the original master, the master right needs attention; if it reuses protected melody or lyrics, the publishing side does too.
Under U.S. law, the copyright owner holds the exclusive right to prepare derivative works — works in which preexisting material is recast, transformed or adapted — and the Copyright Office is explicit that unlawfully used preexisting material doesn't become protected just because new work was built around it. Fair use isn't a default license for remixes; it's a fact-specific doctrine. Pitch-shifting the vocal, chopping it into slivers or burying it under effects doesn't automatically make the use lawful. If the whole appeal of your new track is that listeners recognize the extracted performance, treat clearance as a central production task, not an afterthought.
A hypothetical release that goes wrong
A producer separates the vocal from an old dance record and drops two lines into a new beat. The output has some artifacts, so the producer figures it's changed enough to be safe. The distributor accepts the track, it goes live, and it starts showing up in short-form videos.
Then the original label files a master-rights complaint. The publisher raises a separate claim, because the reused lines carry the original melody and lyrics. The distributor pauses the release and asks for licenses. The producer sends screenshots from the stem-separation software.
Those screenshots prove how the audio was extracted. They don't prove permission. A technically transformed file can still reproduce protected material, and delivery to platforms never validates the rights behind it.
A new version may need a new ISRC
An extracted instrumental, a remix or a materially different edit shouldn't just inherit the original recording's identifier. An ISRC identifies a particular sound recording, not the underlying composition — and IFPI's guidance generally treats remixes and materially different edits as new recordings that need new ISRCs, while a straightforward remaster can keep the existing code when it stays fundamentally the same recording.
Before delivery, compare the new file against the existing master:
- Has the performance been rearranged?
- Were elements removed or replaced?
- Is it presented as an instrumental, remix or alternate version?
- Does it have a new duration?
- Is there meaningful new creative input?
- Should the original stay separately available?
Use accurate version information in the title and metadata, and preserve the original release date for the original master — don't make a freshly separated catalog version masquerade as the historical recording. Artists preparing an authorized alternate version through CREWPORT can use its guided delivery and metadata checks, including assignment of a new ISRC where appropriate. The rights decision still belongs to the artist, label and relevant rightsholders — the tooling handles the how, not the may I.
Keep a stem-separation record
For every commercial project involving separated audio, keep: the untouched source file; evidence of who controls the source master; composition ownership and publishing info; remix, sample, performer and featured-artist approvals; the software and version; the date and processing settings; the original separated outputs; the final edited stems and master; notes on restoration or production changes; the identifier and version metadata for the new recording; any territory, term, media or promo restrictions; and the delivered release confirmation.
One more, easy to overlook: for confidential or unreleased material, find out whether the separation runs locally or uploads your audio to a remote service. Ableton states its built-in process runs locally; other tools may use cloud processing under different privacy, retention and training terms. Don't upload a client's unreleased master to an external service just because the interface offers a free trial.
Separation recovers possibilities, not permissions
Stem separation can rescue a vocal from a cassette, build an instrumental from a lost session, and reopen a catalog that looked technically frozen. It can also make unauthorized sampling feel deceptively clean.
The boundary is simple. If you control the material, the tool may recover options you thought were gone. If somebody else controls it, the ease of extraction changes nothing about the need for permission. A decade ago, losing the multitracks could end a project; today it may just change the workflow. That's genuine progress — just don't confuse access to the sound with ownership of what you found inside it.
FAQ
Does AI stem separation give me the original studio stems?
No. It estimates and reconstructs approximations from the finished stereo mix — it doesn't reopen the original multitrack. Expect some artifacts (reverb residue, phasing, missing consonants), and always compare against the untouched source.
Can I remix or sample a song just because my DAW can separate it?
No. Being able to extract a stem isn't permission to use it. Sampling, distributing, monetizing, performing or syncing extracted audio from someone else's master implicates the master rights, and reused melody or lyrics implicate the composition.
Does an "AI stem" count as new, royalty-free audio?
No. If it contains sound taken from an existing recording, it still carries that recording's protected material. The AI process doesn't strip the copyright out of it.
If I own my master, can I do anything I want with the stems?
Mostly, on the recording side — but check contracts. Featured-performer consent, session-musician terms, label approval clauses or composition rights (if you change the underlying song) can still apply.
Do I need a new ISRC for a separated instrumental or remix?
Usually yes. IFPI generally treats remixes and materially different edits as new recordings needing new ISRCs; a straightforward remaster can keep the original. Use accurate version metadata and keep the original's release date intact.
Is my audio uploaded when I separate stems?
Depends on the tool. Ableton's built-in separation runs locally; other tools may process in the cloud under different privacy and retention terms. Don't upload confidential or unreleased masters without checking.
Release your alternate versions cleanly
Once the rights are sorted, the delivery should be the easy part. An authorized instrumental, remix or alternate edit needs accurate version metadata and, usually, its own ISRC — so it doesn't get mistaken for the original master.
CREWPORT offers guided delivery and metadata checks, including a new ISRC where appropriate, and keeps your version info and original release dates straight across your catalog. The rights call stays yours; CREWPORT makes the clean release simple.
This article is general information, not legal advice. Copyright and contract rules are fact-specific — confirm permissions with the relevant rightsholders and get qualified advice before releasing anything built on separated audio.
Sources
- Ableton — Stem Separation Reference Manual
- Ableton — Live 12.4 Stem Separation Updates
- Ableton — Push Release Notes, July 2026
- Apple — Logic Pro Stem Splitter Guide
- Steinberg — Cubase 15 Stem Separation
- The Beatles — The Making of "Now and Then"
- U.S. Copyright Office — Musical Compositions and Sound Recordings
- U.S. Copyright Office — Derivative Works and Preexisting Material
- IFPI — ISRC Frequently Asked Questions