UMG and ElevenLabs announced personalized vocal experiences for participating artists on September 10, 2026, but neither company promised unrestricted artist voice cloning. The phrase appears beside remixes, mashups, and new track interpretations, with no technical description of how a fan would create or control those vocals.
Artist participation is the important boundary. Artistparticipation in UMG’s ElevenLabs deal is part of the platform’s foundation, not an optional extra added after launch. No participating singer has been named publicly, and no artist voice catalog has been announced.
“Personalized vocals” also leaves several possibilities open. A feature could alter a fan’s own vocal, let an artist offer an approved model, reshape an existing licensed performance, or provide a tightly controlled artist-specific experience. None of those possibilities has been confirmed for the UMG platform.
A person who wants another ElevenLabs user to access their professional clone has to create and verify it on their own account first. They can then share the verified voice privately or make an eligible clone available through the Voice Library. Control begins with the speaker, rather than with the person who happens to possess recordings of them.
The existing product has another limitation that matters more for music. ElevenLabs says Professional Voice Clones do not currently support singing, and the uploaded training material must contain spoken voice. A fan cannot look at the ordinary Professional Voice Clone workflow and assume it is the engine behind a future UMG singing feature.
ElevenLabs could build different technology for the new platform. The companies are jointly developing AI audio products, and the UMG service will be offered separately from ElevenMusic and the Music API. A separate product can have different models, verification steps, permissions, and controls.
Current reporting mostly stops at the phrase “personalized vocals.” The existing voice product shows why the phrase should not be casually rewritten as “clone your favorite singer.” ElevenLabs already distinguishes between making a voice model, verifying who owns the voice, sharing access to it, and determining what the model can actually perform.
The wording in that announcement was much more specific. UMG said artists would retain control over ownership and receive artistic approval and control over output. It also said the custom vocal models were intended for artists’ exclusive creative use rather than general public access.
MicDrop was described as supporting voice-to-voice transformation, speech-to-singing, language changes, and other vocal manipulation. UMG therefore has no difficulty using direct language when a partnership involves artist-trained vocal models. The ElevenLabs announcement does not use the same terminology.
The difference does not prove ElevenLabs will avoid artist voice models. It does show that “personalized vocal experiences” carries less information than “official vocal model trained on an artist’s own voice data.” Treating those phrases as interchangeable erases a meaningful gap in what has actually been disclosed.
Recent research on cloned vocal identity rights examines this problem across several legal frameworks and treats AI voice cloning as a distinct identity issue. The practical point is simple. Clearing a song does not automatically answer every question about generating a new performance that sounds like a real person.
UMG has taken the same distinction seriously in public statements. The company has backed legal protection against unauthorized AI replicas of a person’s voice or likeness and has repeatedly framed responsible AI around artist authorization and control.
A fan-facing platform therefore needs more than a broad music license if it lets users generate recognizable new artist vocals. The product would need rules covering which artists participate, which vocal uses they approve, how outputs are constrained, and what happens when an artist withdraws or changes permission.
None of those controls have been published for the ElevenLabs platform yet. Until they are, the safest description is exactly the one the companies chose. Fans are being promised personalized vocal experiences involving participating artists, not an open celebrity voice-cloning machine.
Artist participation is the important boundary. Artistparticipation in UMG’s ElevenLabs deal is part of the platform’s foundation, not an optional extra added after launch. No participating singer has been named publicly, and no artist voice catalog has been announced.
“Personalized vocals” also leaves several possibilities open. A feature could alter a fan’s own vocal, let an artist offer an approved model, reshape an existing licensed performance, or provide a tightly controlled artist-specific experience. None of those possibilities has been confirmed for the UMG platform.
ElevenLabs does not normally let you clone another person
ElevenLabs’ current Professional Voice Clone product has a hard identity check. You can create a professional clone of your own voice, but the company says you cannot create one of somebody else’s voice, even when that person has permitted you.A person who wants another ElevenLabs user to access their professional clone has to create and verify it on their own account first. They can then share the verified voice privately or make an eligible clone available through the Voice Library. Control begins with the speaker, rather than with the person who happens to possess recordings of them.
The existing product has another limitation that matters more for music. ElevenLabs says Professional Voice Clones do not currently support singing, and the uploaded training material must contain spoken voice. A fan cannot look at the ordinary Professional Voice Clone workflow and assume it is the engine behind a future UMG singing feature.
ElevenLabs could build different technology for the new platform. The companies are jointly developing AI audio products, and the UMG service will be offered separately from ElevenMusic and the Music API. A separate product can have different models, verification steps, permissions, and controls.
Current reporting mostly stops at the phrase “personalized vocals.” The existing voice product shows why the phrase should not be casually rewritten as “clone your favorite singer.” ElevenLabs already distinguishes between making a voice model, verifying who owns the voice, sharing access to it, and determining what the model can actually perform.
UMG has used clearer language for real vocal models
UMG has previously announced an AI deal that really did involve artist vocal modeling. Its 2024 agreement with SoundLabs gave UMG artists access to MicDrop, a tool designed to create official high-fidelity vocal models using the artists’ own voice data.The wording in that announcement was much more specific. UMG said artists would retain control over ownership and receive artistic approval and control over output. It also said the custom vocal models were intended for artists’ exclusive creative use rather than general public access.
MicDrop was described as supporting voice-to-voice transformation, speech-to-singing, language changes, and other vocal manipulation. UMG therefore has no difficulty using direct language when a partnership involves artist-trained vocal models. The ElevenLabs announcement does not use the same terminology.
The difference does not prove ElevenLabs will avoid artist voice models. It does show that “personalized vocal experiences” carries less information than “official vocal model trained on an artist’s own voice data.” Treating those phrases as interchangeable erases a meaningful gap in what has actually been disclosed.
Voice permission can sit outside ordinary song licensing
A singer’s recognizable voice creates issues beyond permission to use a recording or composition. Legal systems can protect vocal identity through publicity, personality, privacy, or related rights even when a dispute does not fit neatly inside traditional copyright.Recent research on cloned vocal identity rights examines this problem across several legal frameworks and treats AI voice cloning as a distinct identity issue. The practical point is simple. Clearing a song does not automatically answer every question about generating a new performance that sounds like a real person.
UMG has taken the same distinction seriously in public statements. The company has backed legal protection against unauthorized AI replicas of a person’s voice or likeness and has repeatedly framed responsible AI around artist authorization and control.
A fan-facing platform therefore needs more than a broad music license if it lets users generate recognizable new artist vocals. The product would need rules covering which artists participate, which vocal uses they approve, how outputs are constrained, and what happens when an artist withdraws or changes permission.
None of those controls have been published for the ElevenLabs platform yet. Until they are, the safest description is exactly the one the companies chose. Fans are being promised personalized vocal experiences involving participating artists, not an open celebrity voice-cloning machine.
