Protecting Human Voices in the Age of AI: Australian Copyright Law and Beyond

In the lead-up to Australia’s 2025 federal election, residents of Burnie, a coastal city in Tasmania (Australia), were asked to listen to a recording of Senator Jacqui Lambie outlining a re-election promise to relocate the National Cabinet from Canberra to Tasmania. Although many responded favourably to Senator Lambie, there was a catch: the voice they heard was not Lambie’s but rather an AI-generated voice clone. This incident is one of a growing number of examples illustrating the increasing prevalence of AI-generated ‘deepvoices’ in modern society.

Deepvoice technology offers exciting possibilities. Films can be dubbed into multiple languages using the original actor’s voice, the speech of deceased persons can be recreated for documentaries, and song parodies can be produced to sound convincingly like well-known artists. Commercially, marketing campaigns can also be maximised, with the potential for a single voice recording to be deployed across multiple projects in an infinite number of ways.

However, these opportunities are accompanied by significant risks. Unauthorised voice cloning raises fundamental questions about identity, consent, and trust. A convincing deepvoice can be used to spread misinformation, facilitate fraud, or damage reputations. Given that every individual has a voice, no one is immune to these harms.

At a professional level, deepvoice technology also poses risks for voice actors, musicians and other performers whose income is derived from their voice. Earning opportunities may be diminished as AI-generated voice clones can be created quickly and at relatively low cost, potentially displacing human performers.

The widespread implications of this development raise a critical question: do current laws adequately protect the human voice? This post examines intellectual property law in two jurisdictions, Australia and the US, to compare each legal system’s approach to this emerging technology.

Voice protection in Australia

While the human voice itself is not recognised as a protected subject matter under Australian copyright law, Part IV of the Copyright Act 1968 protects sound recordings. As a result, copyright is likely to subsist in the recorded voice material that forms part of the datasets used to train deepvoice systems.

However, this copyright protection offers only limited safeguard at the ‘input’, or training stage. Due to territoriality of copyright law, Australian copyright can only be enforced when the infringing use of the recordings occurs within Australia. This creates a significant jurisdictional limitation, particularly since to date, training of generative AI models largely takes place overseas. Even where jurisdiction is not an obstacle, the lack of transparency surrounding AI training datasets often makes it difficult, if not impossible, for individuals to determine which specific recordings of their voices have been used to train a model, as required to bring an infringement claim.

Importantly, copyright in voice recordings is frequently owned by producers or other intermediaries rather than the individuals whose voices were recorded. Consequently, even where a recording is used without authorisation, the person most directly affected may not have a right to personally enforce it.

At the ‘output’ stage where the voice clones are generated, copyright protection remains difficult. AI-generated voice clones are not exact replicas of underlying voice recordings, but rather novel outputs based on learned patterns from large datasets. As such, the element of ‘substantial similarity’ for copyright infringement is unlikely to be satisfied, as it requires any reproduction to be an actual copy of the original recording and not merely a soundalike version.

Beyond a copyright framework, Australia’s passing off law provides limited utility. This doctrine is concerned with protecting commercial reputation from misrepresentation. It requires proof of established reputation, consumer confusion, and financial damage. While this may help well-known figures whose synthetic voice is used in a way that confuses consumers, it offers little protection for ordinary individuals or less prominent professionals. In addition, a passing off claim is unlikely to succeed if the deepvoice is not misleading, such as when a disclaimer is provided to indicate that the voice was synthetically generated.

Voice protection in the US

In the US, the broad copyright fair use exception affords even less protection against the unauthorised use of human voice in deepvoice training. However, other legal avenues are available for individuals to protect their voice from AI misuse.

The US ‘right of publicity’ allows individuals to control the commercial use of their identity, including their voice. This provides a more direct route for individuals to challenge AI voice replication compared to copyright law since voice itself is protected, rather than only underlying recordings. Presently however, this right varies between states and does not always address unique technological challenges.

To bridge these gaps, some US jurisdictions are introducing targeted legislation, such as Tennessee’s ELVIS Act and the proposed federal NO FAKES Act. These laws explicitly protect voices against AI replication and give individuals control over AI-generated versions of their voice and likeness.

Where next?

Against the backdrop of Australia’s limited legal voice protection and the more proactive approach emerging in the US, the question becomes whether jurisdictions without specific voice protection, like Australia, should now consider introducing a dedicated ‘right of voice’. By giving individuals greater control over their voice, such a right could address the harms posed by unauthorised deepvoice creation and distribution. At the same time, it could define space for legitimate innovation by clarifying lawful uses, thereby balancing protection and creativity.

On the other hand, several challenges would need to be addressed. AI industry stakeholders are likely to argue that introducing a restrictive right could stifle AI innovation and impose significant compliance burdens on developers. More fundamentally, it remains unclear whether such a right could be integrated into legal systems, such as Australia’s, that do not have a well-established tradition of personality protections.

For a more detailed analysis of this question please read our full paper: The Protection of Human Voice in the Age of AI by Tamika Mansell, Rita Matulionyte :: SSRN

Photo by engin akyurt on Unsplash