A synthetic vocal can now reproduce the tone, accent and phrasing of a recognisable performer without that person entering a recording studio.

That technology can support legitimate creative work. An artist may authorise a digital version of their own voice for a specific project, an approved alternate performance or a controlled production experiment.

The same technology can also be used to release music that falsely appears to feature a real singer, imitate a deceased performer or make an artist seem to endorse lyrics and products they never approved.

The difference is not the quality of the technology.

It is consent, control, transparency and responsibility.

Voice cloning becomes especially complicated in music because a single track may involve separate rights in the composition, sound recording, performance, artist identity and generated output. Labelling a recording as AI-generated may inform listeners, but it does not automatically make the use lawful.

What Is AI Voice Cloning?

Voice cloning uses digital technology to create an artificial model capable of reproducing characteristics associated with a particular human voice.

The result may imitate:

• vocal tone;
• pronunciation;
• accent;
• pitch range;
• timing;
• breath patterns;
• phrasing;
• expressive delivery.

SAG-AFTRA defines a digital replica as a replica of a person’s voice or likeness created using digital technology such as artificial intelligence. The U.S. Copyright Office uses the wider concept of a digital replica to describe highly realistic technology-generated representations of an individual’s voice or appearance.

A cloned voice should be distinguished from ordinary vocal processing.

Traditional production tools may:

• tune pitch;
• adjust timing;
• change formants;
• remove noise;
• double a performance;
• alter tone;
• isolate an existing vocal.

These processes modify an actual recording.

A voice clone can create a new performance that the person never recorded.

Voice Cloning Is Not the Same as Vocal Editing

The difference matters creatively and legally.

A vocalist may approve an engineer correcting the tuning of a recorded performance. That does not necessarily mean the vocalist approved the creation of a reusable digital model capable of singing new lyrics.

California’s digital-replica contract law expressly distinguishes highly realistic computer-generated replicas from authorised practices such as sampling, remixing, mastering and digital remastering. SAG-AFTRA’s Sound Recordings Code similarly separates digital voice replicas from established production technologies used to alter pitch, timing, timbre, tempo or other properties of an existing recording.

That creates a practical boundary:

• Vocal editing changes a performance the artist actually made.
• Voice cloning can manufacture a new performance attributed to that artist’s vocal identity.

A contract granting permission to edit a vocal should not be treated automatically as permission to generate new vocals.

Why Voice Cloning Is Different in Music

A singer’s voice is not merely a technical input.

It may be central to:

• public identity;
• artistic reputation;
• contractual exclusivity;
• genre positioning;
• fan recognition;
• endorsement value;
• future earning potential.

An unauthorised synthetic vocal may make the public believe that an artist:

• recorded the song;
• approved its lyrics;
• collaborated with another artist;
• endorsed a political message;
• promoted a product;
• returned after retirement;
• participated in a posthumous release.

The potential harm can exist even when no section of an existing sound recording was copied directly.

The identity of the performer may be the thing being appropriated.

Copyright Is Only One Part of the Legal Analysis

Voice-cloning disputes are often described as copyright problems, but copyright does not answer every question.

The U.S. Copyright Office concluded that digital replicas do not fit neatly within one existing area of law. Relevant protections may involve copyright, state publicity and privacy rights, contracts, unfair competition, fraud, consumer protection and other legal frameworks. The Office described current state protections as inconsistent and dependent partly on where the affected person lives or where the conduct occurs.

An artist reviewing a synthetic vocal may therefore need to ask several separate questions.

Was an existing recording copied?

If a system or producer copied protected audio, samples, lyrics or music, copyright may be directly relevant.

Was the performer’s identity replicated?

A newly generated vocal may imitate a singer without reproducing the exact sounds from an existing recording. That can raise voice, likeness, publicity or digital-replica issues.

Was the use authorised by contract?

A label, producer or technology company may claim permission under an existing agreement. The wording, specificity and enforceability of that agreement will matter.

Were listeners misled?

Marketing a synthetic performance as an authentic artist collaboration may raise separate questions involving misrepresentation or false endorsement.

A Voice Is Not the Same as a Musical Composition

A song may contain multiple separate rights.

The underlying composition can include:

• melody;
• harmony;
• lyrics;
• musical structure.

The sound recording is the particular recorded performance.

The performer’s recognisable voice is another interest that may be protected through identity, publicity, contract or digital-replica rules rather than ordinary ownership of the composition.

This means a producer could theoretically generate:

• new lyrics;
• a new melody;
• a new instrumental;
• a cloned vocal resembling a famous singer.

Even when the new song does not copy an existing composition, the synthetic vocal can still create a serious identity and consent issue.

The United States Still Has a State-by-State Dimension

The U.S. Copyright Office has recommended a federal right protecting individuals against the knowing distribution of unauthorised digital replicas. It concluded that existing laws do not provide sufficient and consistent redress and that a nationwide baseline is needed while allowing states to maintain additional protections.

State laws already address parts of the problem in different ways.

Tennessee’s ELVIS Act

Tennessee enacted the Ensuring Likeness, Voice, and Image Security Act, commonly known as the ELVIS Act.

The legislation added voice to the personal rights protected under Tennessee law and was promoted specifically as a response to unauthorised AI uses affecting songwriters, performers and music professionals.

The law is significant for music because it recognises that an identifiable voice can hold value independently of one particular recording.

However, artists should not assume that one state law governs every international release or every online use. Applicable rights may depend on the people, contracts, conduct and jurisdictions involved.

California’s Contract Protections

California Labor Code Section 927, effective for qualifying new performances fixed on or after January 1, 2025, addresses certain contract provisions involving digital replicas.

A provision can be unenforceable as it relates to a new digital-replica performance when it permits the replica to replace work the individual would otherwise have performed, lacks a reasonably specific description of the intended uses and the individual was not represented by legal counsel or an applicable labour union.

The law does not mean that every digital-replica licence in California is invalid.

It shows why vague language such as “all technologies now known or later developed” may be inadequate when a performer is being asked to authorise a synthetic replacement.

Digital Replicas of Deceased Performers

The legal and ethical questions do not disappear when the performer has died.

California law can impose liability for producing, distributing or making available certain unauthorised digital replicas of a deceased personality’s voice or likeness in an expressive audiovisual work or sound recording. The law also contains exceptions for uses including news, criticism, scholarship, satire, parody and qualifying historical or biographical treatment.

A deceased artist’s catalogue may also involve:

• estate rights;
• label contracts;
• master ownership;
• publishing rights;
• performer agreements;
• prior lifetime licences;
• moral and reputational interests.

Permission from the owner of an old master recording does not necessarily settle every question about manufacturing a completely new vocal performance in the deceased artist’s voice.

Union Agreements Provide a Practical Consent Model

The 2024 SAG-AFTRA Sound Recordings Code contains specific digital-voice-replica provisions for covered sound recordings produced under the agreement.

It requires consent before releasing a covered recording featuring a digital replica of the artist’s voice. The consent must be clear, conspicuous, contained in a separate writing and based on a reasonably specific description of the intended use. Newly provided consent is generally required on a project basis.

The agreement also includes compensation provisions. It states that digital-replica use in a released covered recording must be compensated according to the applicable artist agreement for royalty artists and under specified minimum terms for non-royalty performers.

These rules do not govern every independent recording.

They provide a useful professional framework built around:

1. informed consent;
2. defined use;
3. compensation;
4. project-specific control.

Consent Must Be Specific

A performer cannot make an informed decision when the proposed use is described only as “AI purposes.”

Useful consent should identify:

• who will create the voice model;
• which source recordings will be used;
• which project will use the replica;
• whether new lyrics can be generated;
• whether commercial release is planned;
• the permitted territory;
• the licence duration;
• the platforms involved;
• whether sublicensing is allowed;
• whether the model can be reused;
• whether the performer can review the result;
• how compensation will work;
• how the model and source data will be stored;
• what happens after the licence ends.

Consent to one track should not silently become consent to an unlimited catalogue.

Consent to a private demonstration should not automatically become consent to public distribution.

The Performer Should Control the Context

The same synthetic voice can create very different implications depending on the content.

A vocalist may accept a cloned performance for:

• an approved harmony layer;
• a fictional character;
• a technical demonstration;
• an alternate language project;
• an authorised archival production.

The performer may reject its use for:

• political advertising;
• explicit lyrics;
• gambling promotion;
• deceptive collaborations;
• defamatory material;
• competing artist projects;
• products that conflict with existing endorsements.

SAG-AFTRA agreements in several AI contexts emphasise informed consent and control over particular uses rather than a one-time unrestricted approval.

For independent artists, the same principle should apply even without union coverage.

Legitimate Creative Uses of Voice Cloning

Voice-cloning technology is not inherently abusive.

A controlled project may use it to:

• create an approved synthetic instrument from the artist’s own voice;
• explore harmonies beyond the singer’s comfortable range;
• produce a clearly disclosed fictional character;
• assist a performer with accessibility needs;
• create authorised variations for an interactive project;
• restore a limited missing element where appropriate rights exist;
• build experimental live or installation work;
• preserve an artist-approved vocal model under defined conditions.

The legitimacy of the project depends on more than creative intention.

It requires agreement on:

• source material;
• rights;
• consent;
• credits;
• data storage;
• compensation;
• release authority.

A technically interesting use can still be contractually or ethically unacceptable.

Voice Restoration and Voice Replacement Are Different

AI can support restoration without necessarily creating a complete replica.

For example, a tool may attempt to:

• reduce noise;
• reconstruct a damaged syllable;
• isolate a vocal;
• repair clipping;
• extend a short sound.

The closer the output remains to repairing the artist’s actual performance, the more it resembles advanced editing.

The more the system invents new words, phrasing or emotional delivery, the closer it moves toward a synthetic performance.

That boundary should be documented.

The final credit should not imply that the artist personally recorded material they never performed unless the project and artist have expressly approved that representation.

Voice Cloning Can Affect Employment

A digital replica may allow a company to generate future performances without rehiring the vocalist.

That is one reason compensation and use limitations matter.

A one-hour modelling session could potentially create an asset capable of producing:

• multiple tracks;
• demos;
• advertisements;
• live content;
• translations;
• personalised outputs.

The payment should reflect the permitted use rather than only the duration of the original recording session.

California’s contract rules specifically address situations where a digital replica is used in place of work the individual would otherwise have performed. SAG-AFTRA’s sound-recording provisions also require compensation for covered released uses.

Artist Style and Artist Voice Are Not Identical

A prompt requesting “a sad 1980s synth-pop vocal” describes a broad creative direction.

A prompt requesting the voice of a named living singer targets an identifiable individual.

The U.S. Copyright Office distinguished concerns about artistic style from highly realistic replicas of a specific person. While recognising concerns about style imitation, it recommended that new digital-replica protection focus on representations that convincingly appear to be the actual individual.

That does not make style imitation risk-free.

A generated track may still copy:

• lyrics;
• melody;
• a protected recording;
• artwork;
• branding;
• misleading artist attribution.

But a general stylistic influence and a realistic clone of a named voice should not be treated as the same legal category.

Disclosure Does Not Replace Consent

A label stating “AI-generated vocal” can help listeners understand the production process.

It cannot authorise the use of another person’s identity.

An unauthorised recording does not become legitimate because:

• AI use is disclosed;
• the track is described as a tribute;
• the creator did not charge money;
• the artist’s name is omitted from the title;
• a disclaimer states that the singer was not involved.

Disclosure may reduce confusion. It does not automatically resolve publicity, contract, copyright or other rights.

How Voice Cloning Can Become Identity Theft

The phrase identity theft is sometimes used broadly rather than as a precise legal classification.

In practical music-industry terms, a synthetic voice begins to resemble identity exploitation when it is used to make others believe that a real performer said, sang or approved something they did not.

Warning signs include:

• using the artist’s name in the release title;
• presenting the track on an incorrect artist profile;
• copying official artwork or branding;
• describing the song as an unreleased collaboration;
• generating fabricated interviews;
• using the voice to request money or access;
• impersonating the artist in private communications;
• hiding the synthetic nature of the performance.

The harm can extend beyond one release.

Listeners may stop trusting authentic recordings, official announcements and legitimate artist communications.

Risks for Independent Artists Using Their Own Voice

An artist can also lose control of a model they voluntarily created.

Before providing source recordings, check:

• whether the company retains them;
• whether the model is shared;
• whether the artist receives a private or public model;
• whether other users can generate outputs;
• whether the company may use the data for training;
• whether deletion is technically and contractually available;
• whether the model can be transferred if the company is sold;
• whether the artist can revoke future uses;
• whether security standards are documented.

A voice model can be more sensitive than an ordinary audio file because it may be reusable.

Deleting the original upload may not necessarily remove every model, backup or derived asset unless the agreement addresses those elements.

What a Voice-Clone Agreement Should Cover

A professional agreement should define the specific relationship rather than relying on a generic release form.

Identity of the model

State whose voice is being replicated and which source recordings are authorised.

Permitted project

Identify the song, album, advertisement, game, installation or other approved project.

Permitted outputs

Define whether the model may generate:

• singing;
• speech;
• harmonies;
• translations;
• live outputs;
• personalised versions.

Approval

State whether the performer can review:

• lyrics;
• final audio;
• artwork;
• marketing;
• credits;
• future variations.

Compensation

Clarify:

• creation fee;
• session fee;
• reuse fee;
• royalty;
• revenue share;
• minimum payment;
• additional-project fees.

Duration and territory

Avoid an unlimited licence unless the performer fully understands and intentionally accepts it.

Storage and security

Specify who stores:

• raw voice data;
• model weights;
• generated files;
• access credentials.

Revocation and termination

Define what happens when:

• the project ends;
• the company breaches the agreement;
• the performer dies;
• the model is compromised;
• the licence expires.

Credits and disclosure

Clarify how the synthetic performance will be described publicly.

What Labels and Distributors Should Verify

A label or distributor receiving a synthetic-vocal recording should not rely solely on a verbal statement that “the artist approved it.”

The delivery file should include:

• performer identity;
• written consent;
• scope of permitted use;
• voice-model provider;
• commercial-use rights;
• release authority;
• required credit;
• AI classification;
• restrictions on territory or platform;
• estate or representative approval where applicable.

Where the vocal resembles a known artist who is not credited as a participant, the release deserves additional review.

What Studios and Producers Should Document

A studio should maintain a clear distinction between:

• the original human recording;
• processed versions;
• model-training files;
• generated vocal outputs;
• approved final takes.

The project archive should record:

• who requested the clone;
• who authorised it;
• which tool was used;
• what data was uploaded;
• who had access;
• which outputs were delivered;
• what was deleted;
• what remained in the master.

The studio should not create a reusable voice model merely because it has access to isolated vocals from an earlier session.

Possession of the recording is not the same as permission for a new use.

Voice-Cloning Risk Checklist

Before creating or releasing a synthetic vocal, confirm:

• The performer is clearly identified.
• Written consent exists.
• The intended use is specific.
• The performer knows which lyrics will be generated.
• Compensation is documented.
• The tool’s commercial terms have been reviewed.
• Source recordings were lawfully supplied.
• The model is stored securely.
• Reuse rights are limited.
• Collaborators know the vocal is synthetic.
• The artist profile will not be misused.
• Credits are accurate.
• AI disclosure is prepared.
• The output does not copy protected lyrics or melodies.
• Deceased-performer rights have been reviewed.
• The performer or representative can report misuse.
• Termination and deletion procedures are defined.

What to Do if Your Voice Has Been Cloned

A performer who discovers a suspected unauthorised clone should preserve evidence before links disappear.

Useful records may include:

• exact URLs;
• screenshots;
• audio files;
• release metadata;
• artist-profile pages;
• uploader information;
• dates;
• advertisements;
• public statements;
• listener comments showing confusion.

The performer or representative may then need to contact:

• the platform;
• distributor;
• hosting provider;
• social network;
• relevant rights holder;
• union;
• qualified legal adviser.

The appropriate procedure depends on whether the issue involves:

• copyright infringement;
• artist impersonation;
• publicity rights;
• contract breach;
• fraud;
• false attribution;
• platform policy.

A standard copyright takedown may not always be the correct or only route when the central issue is a newly generated imitation of the voice.

Frequently Asked Questions

Is AI voice cloning legal?

Voice cloning is not automatically lawful or unlawful in every situation. Legality can depend on consent, contracts, publicity and privacy rights, copied material, marketing, jurisdiction and the specific use.

Does a singer own the copyright in their voice?

A person’s voice is not identical to copyright in a musical composition or sound recording. Protection against imitation may arise through other laws and contracts, including publicity, privacy and digital-replica rules.

Can a producer clone a vocalist who agreed to record a song?

Permission to record and edit a performance should not automatically be treated as permission to create a reusable digital voice replica. Specific consent should address creation, use, compensation and storage.

Is Auto-Tune voice cloning?

Ordinary pitch correction modifies an existing performance and is generally distinguishable from generating a new performance through a digital voice replica.

Can an artist legally clone their own voice?

An artist can authorise use of their own vocal identity, but they should still review model-provider terms, collaborator agreements, label restrictions, data storage and commercial-use rights.

Can the voice of a deceased singer be cloned?

A project may require approval from an estate, rights holder or authorised representative. California law expressly addresses certain unauthorised digital replicas of deceased personalities in sound recordings and audiovisual works.

Does marking a song AI-generated solve the rights issue?

No. Disclosure can inform listeners, but it does not replace performer consent, copyright clearance, contractual authority or commercial-use permission.

Should a cloned singer receive royalties?

Compensation depends on the governing contract or collective agreement. The SAG-AFTRA Sound Recordings Code includes compensation provisions for covered uses of digital voice replicas.

Final Perspective

Voice cloning can function as a creative instrument, but it is an instrument built from a person’s identity.

That makes it fundamentally different from an ordinary synthesiser preset.

A vocal replica can preserve recognisable characteristics that took an artist years to develop. It can also create performances, statements and associations that the artist never approved.

Responsible use requires more than technical ability.

It requires:

• specific consent;
• defined scope;
• fair compensation;
• secure storage;
• honest credits;
• transparent disclosure;
• meaningful performer control.

The technology itself does not determine whether a project is innovative or exploitative.

The relationship with the person behind the voice does.