Voice As Persona - Legal Implications Of AI- Generated Vocal Cloning In The Music Industry
Author: LawSphereX
Category: Intellectual Property
Published:
Updated:
VOICE AS PERSONA: LEGAL IMPLICATIONS OF AI-
GENERATED VOCAL CLONING IN THE MUSIC INDUSTRY
Mr Kaustav Chowdhury, LL.M., KIIT School of Law, Bhubaneswar
ABSTRACT
The human voice is an essential aspect of both individuality and creative identity. The recent developments in the realm of generative AI technology have facilitated realistic voice cloning and resulted in the emergence of numerous AI-generated vocal covers. The mentioned situation demonstrates important weaknesses of the current copyright, personality rights, publicity rights, and performers' rights law. The concept of “voice as persona” is developed in this article to examine the legal consequences of AI vocal impersonation in the context of the music industry.
This study conducts comparative doctrinal research on the literature available between 2024 and 2026 and examines problems such as text-anddata-mining exemptions and audio mining, fragmented right of publicity, insufficient consent and remuneration regimes, identity appropriation, artist unemployment, and the psychological and ethical threats posed by voice creation. The article further asserts that treating voices as protectable persona will be vital for the establishment of sui generis digital replication rights, incorporation of PRAC3 framework (Privacy, Reputation, Accountability, Consent, Credit, Compensation), and other legislative reforms. This will ensure an equal focus on artistic interests and technology advancement within the context of developing laws on artificial intelligence.
Keywords: Voice as Persona, AI Vocal Cloning, Right of Publicity, Digital Voice Replication, PRAC3 Framework, Music Industry Law
1. Introduction
In April 2023, an AI-created piece of music named “Heart on My Sleeve” became viral online. 1 In spite of the fact that none of Drake and The Weekend, two of the biggest stars in the world today, contributed to this piece of art, it successfully cloned their voices and gained immense popularity in matter of days. However, before it was banned due to copyrights infringement, “Heart on My Sleeve” proved one important thing – AI can clone voices of the greatest musicians of all times, making it practically impossible to tell whether a recording was made by an actual artist. 2 Alongside that, new technologies for cloning voices like ElevenLabs, Voicify, Kits. ai, and others have brought highly accurate vocal synthesis into the reach of consumers. 3 Now anyone is capable of creating an “AI cover” of their favorite music in the voice of performers that never recorded the song in question.
What used to be restricted to labs and researchers is now common practice. This all has implications for the fact that the human voice is not just a source of sound. The voice of an individual constitutes an essential part of personal identity, artistic persona, and commercial success in the music industry. 4 A performer’s unique vocal expression becomes something identifiable, emotional, and economically valuable. To many performers, their voice becomes synonymous with their brand and identity. In appropriating their voice without permission, AI technologies effectively hijack the person’s entire persona. Legal instruments have been unable to respond effectively to the challenge posed by these AI techniques. Copyright protects musical works and sound recordings; however, there is no explicit coverage of vocal timbres or vocal identities per se. 5 Right of publicity in the U. S.
provides varying levels of protection across states. 6 In the EU, the fragmented system of personality rights and data protection may be relevant, but only partially. Emerging countries such as India and Indonesia face similar challenges; for instance, performer’s rights and privacy doctrine are insufficiently developed for coping with these emerging technologies. The issues 1 Ben Sisario, Fake Drake Song Written by A. I. Highlights Perils of New Technology, N. Y. Times (Apr. 17, 2023), https://www. nytimes. com.
2 Id.
3 ElevenLabs, https://elevenlabs.io.
4 Jennifer E. Rothman, The Right of Publicity: Privacy Reimagined for a Public World (2018).
5 U.S.C. §§ 102, 106.
6 J. Thomas McCarthy, The Rights of Publicity and Privacy (2d ed. 2024).
become even more complicated due to TDM exemptions, audio mining, and the inclusion of copyright protected recordings in training databases without explicit authorization of performers. It is therefore apparent that the swift growth of AI voice cloning technology points out the flaws inherent within global legal frameworks. More than economic issues such as unemployment, loss of license fees, or any form of business unfairness, there are deep moral and psychological implications attached to voice duplication without authorization. Such an act may not only alter the intentions of the artist but may also cause damage to his reputation as well as interfere with his freedom of expression. 7 The theoretical construct of “voice as persona” has been developed in this paper to examine the legal challenges arising from vocal cloning using AI technology in the music business sector.
As voice is treated here as an extension of identity in addition to being a technological product, this construct serves to link different areas of law, namely copyright, right of publicity, performers' rights, privacy laws, and regulations for regulating AI technologies. The paper relies on comparative doctrinal research of scholarly and legislative literature between 2024-
2026 to identify gaps in the legal system pertaining to the US, the EU, India, and Indonesia.
Recognition of voice as persona that can be protected warrants the introduction of a sui generis digital right of digital reproduction applicable to identity-cloning by AI. It is also suggested that PRAC3, meaning Privacy, Reputation, Accountability, Consent, Credit, and Compensation, be adopted as a normative approach to guide the development of harmonized legislation in this area. The issue under discussion entails finding a balance between two conflicting requirements: protection of artists' identity and commercial interest on one hand, and technological innovation on the other. In the light of continuing development of generative AI, the issue is not about whether voices can be cloned, but how to respond when identities become reproducible.
2. AI Voice Cloning Overview
The artificial voice replication system involves the use of sophisticated AI techniques such as the training of deep-learning models based on extensive datasets of the spoken language or
7 Berne Convention art. 6bis.
singing. 8 Via the diffusion model or transformer networks, the AI system learns pitch, cadence, tone, and style of speaking/singing. 9 The trained model can then replicate speech or singing sounds very similar to the target voice. There are mainly two processes involved in voice replication, namely (1) training process and (2) generating inference. In the training phase, the AI system is fed audio files containing a target voice to learn its characteristics. In the generating inference phase, the input provided by users is converted into the cloned voice. The advancements that were witnessed from 2024 to 2026 led to an improvement in realism and, consequently, resulted in near-perfect impersonations. The commercial services that exist today provide the ability to generate voices for celebrities and musical artists, at times without permission.
10 There also exists open-source technology where one can generate an impersonation based on their own audio input. Both problems can be legally problematic in both situations. The datasets used in training can contain copyrighted music without the authorization of those who hold the rights. In the output stage, even if there is no copyright infringement on the musical composition or recording, an unauthorized identity performance can occur. In this way, AI-based voice cloning presents both layers of a legal question: one concerning data intake and the other concerning synthesized output. In either case, there is a problem with existing intellectual property theories.
3. Right of Publicity
The publicity rights protect people from being commercially exploited through unauthorized use of their name, voice, likeness, or identity.11 In the U.S., the protection evolved largely under state laws, leading to an inconsistent approach. In California and New York, there is statutory protection of the publicity rights, which even extends beyond death. The courts have held that the use of imitation of a distinctive voice can amount to misappropriation. This has been held in the case of Midler v. Ford Motor Company
8 Ian Goodfellow et al., Deep Learning (MIT Press 2016).
9 Ashish Vaswani et al., Attention Is All You Need, 31 NIPS 5998 (2017).
10 ElevenLabs Documentation, https://elevenlabs.io.
11 J. Thomas McCarthy, The Rights of Publicity and Privacy (2d ed. 2024).
(1988), where the Ninth Circuit Court ruled that imitation of Bette Midler’s voice constituted a violation of her publicity rights. 12 The above discussed precedent gives us an idea that vocal cloning made by AI can be covered by the doctrine of publicity. Nonetheless, there are many problems. Firstly, it should be noted that the protection is geographically limited. Secondly, first amendment claims can be used as a defence for some types of use. 13 Thirdly, in order to claim the protection, commercial use should be proved. The problem is that the EU doesn’t have its own single law protecting publicity rights. Protection is granted according to national laws that grant strong personality rights and human dignity. Personality rights are protected in Germany under the Allgemeines Persönlichkeitsrecht. Similarly, personality attributes are protected in France.
India is an example of the country where personality rights have been acknowledged via constitutional privacy case laws and tort law jurisprudence. 14 In fact, it can be shown by reference to ICC Development v. Arvee Enterprises and other endorsement cases. 15 Nevertheless, in terms of voice recognition and its protection as identity, India also lacks explicit acknowledgment thereof. In Indonesia, protection has been provided via civil code and certain intellectual property case law, but there is less explicit recognition of such protection. Across jurisdictions, a lack of harmonization results in lack of adequate legal protection for artists. Finally, artificial intelligence can render a performance without making a consumer believe that it is made by the artist.
4. Copyright Law and Moral Rights
Copyright Law Copyright is designed to protect works of original expression that are recorded in a tangible medium. It does so for musical works and sound recordings but does not cover the vocal style
12 Midler v. Ford Motor Co., 849 F.2d 460 (9th Cir. 1988).
13 White v. Samsung Electronics America, Inc., 971 F.2d 1395 (9th Cir. 1992).
14 ICC Dev. (Int'l) Ltd. v. Arvee Enters., 2003 (26) PTC 245 (Del).
15 Id.
itself, vocal timbre, or the vocalist's identity. Under American copyright law, there would be no infringement if an AI-generated work sounds like Drake's voice because the copyright would reside in the sound recording and the composition would be different. Additional complications are created by training data. The usage of copyrighted recordings for training AI may involve reproductions. Fair use arguments based on transformative nature and lack of market substitution are commonly used. Recent litigation regarding generative AI technologies in the United States indicates that courts have not yet determined whether data scraping is an infringement. Text and data mining is allowed by the DSM Directive of the European Union in some cases, enabling AI training without infringement. 16 This includes opt-out opportunities for rights owners.
Performers may have difficulties exercising their opt-out opportunities because of the opacity of datasets. Both India and Indonesia lack explicit guidelines in their legislation on this matter. The copyright law of India provides for performers’ protection under Section 38. 17 Nevertheless, there is still no consensus whether this provision is applicable to the use of AI in the creation of digital replicas. Indonesia has laws that cover the economic rights of performers. However, it lacks any specific provisions regarding digital reproduction. However, since copyright centres on works and not individuals, the protection of voice personas is inadequate. • Moral Rights and Ethical Considerations Moral rights are those concerned with the personal relationship between the creator of the content and his work, such as the right of attribution and integrity.
18 Civil law countries recognize moral rights as inalienable. Distorted vocal AI performances can affect the integrity aspects of the content. 19
16 Directive (EU) 2019/790, arts. 3–4.
17 Copyright Act, 1957, § 38 (India).
18 Berne Convention art. 6bis.
19 Id.
There are several other considerations outside the doctrinal issues. For example, AI can deliver lyrics that go against the beliefs of an artist, resulting in damage to reputation. The issue of posthumous clones poses more moral questions than ever before. Psychological impacts on the part of artists include loss of control of their performance and even their reputation. There is also the possibility that audiences might feel disappointed about the lack of authenticity. Unethical vocal impersonation may cause misrepresentation of the work, bring negative consequences to one’s reputation, and create confusion among listeners. It is evident that the undermining of the integrity of musical performances raises issues relating to the very essence of art. It seems that from the voice-persona perspective, ethics should be considered by legislation.
The issue of ethical concerns along with economic matters should thus be taken into consideration when implementing legal measures. The combination of moral rights with copyright will allow us to develop a better solution to the problem.
5. Contract & Licensing
Voice contracts form a critical area of regulation for voice rights. Increasingly, recording agreements contain terms related to AI and digital voice simulation.20 Record labels even go the extra mile by drafting contractual clauses on the use of artificial intelligence with regard to their voices. Nevertheless, the bargaining power disparity presents challenges. Unbeknownst to upcoming talent, the voice contracts they sign could give blanket rights to exploit their voices digitally through simulated speech. In a bid to respect contractual freedom, legislative protections should come into place. Industry-wide voice licensing can prove helpful. Voice databases and licensing arrangements will make it possible for AI exploitation to occur without breaching performers' rights. However, such an arrangement is dependent on a degree of accountability.
20 WIPO, Generative AI and Intellectual Property (2024).
According to the voice as persona theory, any contract regarding voice exploitation needs to involve informed consent. The contract alone is insufficient because systematic inequities requires legislative intervention.
6. Privacy & Biometrics
A voiceprint is biometric information that can be used to identify the individual.21 According
to GDPR regulations under the EU, biometric data that identifies an individual is entitled to additional protection. 22 Thus, the processing of biometric data for the purpose of creating an AI model could invoke data protection law requirements, such as obtaining consent. Also, in the United States, there are state regulations like BIPA of Illinois regarding the collection of biometric identifiers, which may include voiceprints. 23 Cases under BIPA suggest that biometric misuse is taken seriously by the courts. The Indian Digital Personal Data Protection Act (2023) and Indonesian Personal Data Protection Law (2022) incorporate consent-based regimes which might apply to voice data as well. 24 Nonetheless, regulatory mechanisms are still in their infancy stage. Current privacy protection regimes emphasize data collection rather than generated synthetic data.
Yet, considering voice as biometric authentication makes the case for regulating AI voice cloning based on consent.
7. Comparative Analysis and Case Studies
• Comparative Analysis
A comparative approach will show that there is considerable fragmentation among different
jurisdictions. While the US provides a strong and yet inconsistent level of publicity rights, the variation across different states within the US is quite dramatic. Both EU and India have good provisions for privacy and personality rights, albeit in an under-developed manner. There is no digital replication right that covers the issue of AI-induced identity cloning in any
21 GDPR art. 4(14).
22 GDPR art. 9(1).
23 740 Ill. Comp. Stat. 14/1–99 (2008).
24 Digital Personal Data Protection Act, No. 22 of 2023 (India).
jurisdiction. The following table summarizes comparative protection across jurisdictions: Dimension United States European Union India Indonesia Emerging through privacy jurisprudence Strong but statedependent Personality rights under national law Limited civil code provisions Publicity Rights Sound recordings protected; voice not Similar approach Performers' economic rights Copyright Protection Performers' rights under Section 38 BIPA and state laws
GDPR
biometric protections
DPDPA (2023)
PDP
Law (2022) Privacy/Biometrics Recognized under copyright law Limited recognition Robust under Berne Convention Recognized but narrow scope Moral Rights AI-Specific Regulation Emerging AI Act (2024) Developing Nascent • Case Studies § Heart on My Sleeve Incident: The example of the song "Heart on My Sleeve," released in April 2023, shows the weakness of enforcement measures that depend on copyright laws. 25 The song impersonated the voices of Drake and The Weekend in such a way that millions of listens were registered before it was removed from the Internet. § AI-Generated Frank Sinatra Covers: Posthumous clone issues are brought to mind by such synthetic Frank Sinatra performances that appear on TikTok and YouTube platforms. These are impressive but pose ethical issues related to post-mortem publicity rights, moral right to integrity, and estate rights of family members.
§ Authorized Voice Licensing Initiatives (2024–2025): Some of the big recording companies have entered into alliances with AI companies to develop voice models that are officially authorized. This is evident from the alliances that have been formed
25 Sisario, supra note 3.
between Universal Music Group and other companies specializing in voice synthesis technology. § ElevenLabs and Consumer Platforms: However, the growth of applications allowing users to clone their voice, including ElevenLabs, Voicify, and Kits. ai, demonstrates how accessible voice cloning has become. Though each company provides rules of use for their products, their execution is difficult, and there are many questions of authorization for voice cloning based on prior audio recordings. § India Celebrity Endorsement Cases: Recognition of personality rights in India can be seen from the ICC Development v. Arvee Enterprises line of cases. 26 Nevertheless, an explicit expansion to cover voice and AI-generated content is not evident. § EU AI Act Implementation: Introduction of EU AI Act, which is to come into force between 2024 and 2026, marks the regulation of AI on a risk basis.
27 Limited or highrisk voice synthesis software will be regulated accordingly. These case studies together reveal that regulated markets can exist, there are difficulties in enforcement regardless of context, and technology is advancing faster than legislation.
8. Policy Recommendations
It is necessary to fill the doctrinal voids that exist under copyright, publicity, and privacy laws through a structured legislative effort to remedy current gaps in regulation. The existing approach to regulation allows for regulatory arbitrage whereby developers can engage in activities in a jurisdiction with poor regulation while disseminating the content internationally. The following recommendations can be made in relation to addressing the issues of artificial intelligence vocal cloning. • Sui Generis Digital Replication Right Current IP law regimes fail to provide adequate protection for vocal identity. Copyright protects the expression, not the voice itself. Similarly, publicity rights are often divided and
26 ICC Dev. (Int'l) Ltd. v. Arvee Enters., 2003 (26) PTC 245 (Del).
27 Regulation (EU) 2024/1689 (Artificial Intelligence Act).
only apply to situations where endorsements are involved. Hence, there is need for sui generis (unique) legislation to fill this void. a) Scope and Definition: The rights of the person should safeguard their "digital vocal identity," which is the sum total of the elements of tone, pitch, rhythm, and style, through which their voice is recognizable. The protection needs to be extended not only for actual duplication but also for any simulation of the voice. b) Duration and Transferability: Following copyright and publicity laws, the right should continue through the life of the person, supplemented by an additional posthumous duration (say, 50-70 years). Most importantly, this right must be split into two components: the moral part must remain non-transferable, while the economic part can be assigned.
In that way, we will ensure that artists never give up rights to their identity under adhesive agreements. c) Exceptions and Limitations: For there to be an effective balancing act between protecting the individual from misuse of their voice and freedom of expression, it is necessary to have certain exceptions within the right that apply to such uses as parody, satire, reporting, and even teaching purposes. These exceptions must not allow for commercialization to the extent that it interferes with the artist’s licensing process. d) Enforcement Mechanisms: Statutory damages would help ease the burden of proof regarding real economic harm, which can prove challenging in cases involving reputation. Furthermore, safe harbour for such platforms could only be granted upon ensuring that there were appropriate takedown policies in place for unlicensed vocal clones.
• Operationalizing the PRAC3 Framework Legislative language alone is not enough without having normative guidance for its implementation. The PRAC3 concept (Privacy, Reputation, Accountability, Consent, Credit, Compensation) provides an all-encompassing approach to business regulation and legislation. a) Privacy (Biometric Safeguards): The voice data is to be categorized as sensitive biometric information under privacy legislations like the GDPR and BIPA. The use of voice data for training artificial intelligence algorithms should only occur through the
acquisition of explicit consent and not by means of implied consent or general terms of use. b) Reputation (Protection Against Distortion): However, artists should have the power to restrict such uses which portray them in an incorrect manner or misuse their voice on behalf of anything against their morals such as political campaigning or obscene language. This can be seen as an application of the artist's moral rights to preserve their good reputation. c) Accountability (Traceability and Watermarking): Misinformation can be mitigated by ensuring that all voice output from AI technology has an imperceptible watermark or metadata (C2PA). All platforms offering this kind of content must keep audit logs for purposes of determining who created such content in case of abuse. Liability must apply even where platforms knowingly offer infringing clones.
d) Consent (Granular Authorization): The systems of consent must be very granular. A musician may allow the use of his or her voice for a particular video game but not for political advertisements. Consent can also be revocable so that artists can retract their consent to create future generations of models. e) Credit (Disclosure and Attribution): Listeners have the right to be informed as to whether the musical performance is created by human beings or through synthesis. A mandatory label such as “Vocal Performance by Artificial Intelligence” must be provided on streaming services and social media sites. Metadata must give credit to the original human vocalist who lent their voice for cloning purposes.
f) Compensation (Equitable Remuneration): The licensing regime must be designed to guarantee that the artists get fair compensation for not only the training process but also for the commercial utilization of the results. Collective Management Organizations can manage blanket licenses for the AI companies and distribute the royalties according to their usage. • Harmonized International Standards The Internet is without borders, while laws have territorial restrictions. In the absence of international cooperation, there will be vulnerabilities due to less regulated jurisdictions.
a) WIPO-Led Treaty or Model Law: WIPO should organize a diplomatic conference to formulate an international treaty regarding "Protection of Performers' Identities in the Digital Age. " As another option, a model law can be provided by WIPO which may serve as guidelines for national legislation in this regard, as done previously by the UNCITRAL for the regulation of electronic commerce. b) Cross-Border Enforcement: Judgment recognition mechanisms are vital. In case the artist wins in court proceedings against the US-based site within the European Union, the ruling must be valid in all jurisdictions where the mechanism is adopted, and thus will lower the cost of litigation. c) Capacity Building for Emerging Markets: Such jurisdictions need technical and legal assistance in enforcing such standards.
There is a need for international organizations to make provision for technology transfer and capacity building for legal institutions dealing with issues related to artificial intelligence. d) Preventing Regulatory Arbitrage: There must be minimum “floor” requirements that ensure corporations cannot incorporate in countries with less stringent laws and regulations to escape the laws in other countries where they operate. Conditions on market access (for example, denial of services by non-compliance) can motivate global compliance. These efforts strive to create a legal environment where innovations in technology will not result in a loss of human dignity. The recognition of voice as persona within the law acknowledges the reality that technology can copy sounds but cannot replicate an identity.
9. Conclusion
The emergence of voice cloning via artificial intelligence presents a paradigm shift in the development of creative technology. Although such innovation brings about new possibilities for creativity, it also poses fundamental challenges regarding issues of authorship, identity, and ownership. Current laws regarding copyright, publicity, privacy, and contractual rights offer limited protection. The concept of “voice as persona” makes it clear that vocal identity extends beyond mere performance and deserves legal recognition as such.
The introduction of a sui generis digital reproduction right together with implementation of the PRAC3 model would ensure that appropriate safeguards are in place for the future. Such consistent reforms are vital for ensuring that advancements in technology do not infringe upon the artistic freedom of expression. The rise of generative AI raises questions about whether the law can deal with the problem posed by identity being replicable in such an environment. The solution lies in acknowledging the link between voice and persona.