For generations of Dragon Ball viewers, Masako Nozawa’s work as Goku has been inseparable from the character. That connection is precisely why unauthorized AI reproductions of an actor’s voice carry stakes beyond a novelty sound clip. Speaking as the Japan Actors Union introduced new generative-AI guidance, Nozawa argued that using a performer’s voice without permission is “no different than theft.”

Her objection is direct: a voice and the performance expressed through it belong to the actor. “My voice is mine, as are my performances,” Nozawa said, adding that anybody wishing to use it, even in a limited capacity, should ask first. She also said an AI version of her voice she had heard lacked “heart.”

The union’s first version of the guidance lays out a practical position on a difficult issue: voice replication should not be treated as raw material that can simply be collected, trained on, and deployed. Consent is required at every stage, and authorized synthetic-voice use should be recognized as a performance that merits pay and/or royalties.

What the Japan Actors Union guidance covers

The Japan Actors Union, which represents roughly 2,600 actors and voice actors, presented the guidelines at a Thursday press conference. They address audio made through AI that replicates actors’ voices, including voice synthesis and uses connected to deep learning and machine learning.

Those terms matter because they describe different points in a system’s lifecycle:

  • Voice synthesis is the generation of new speech that resembles a person’s voice.
  • Machine learning broadly refers to systems finding patterns in data in order to produce outputs or make predictions.
  • Deep learning is a machine-learning approach that uses layered neural networks and is commonly associated with the training of generative systems.

The union’s proposed principle is not limited to the final audio a listener hears. If a company wants to use an actor’s voice in AI training, it must first get that performer’s permission. If the resulting AI voice is used with authorization, the guidelines categorize that use as a performance, with salaries and/or royalties due to the actor.

That distinction is significant. A debate framed only around whether a finished clip is convincing can overlook the earlier decision to use a person’s recordings as training material. The guidance instead treats consent as an ongoing requirement rather than a one-time box to tick after a model has already learned from an actor’s work.

Why a voice is more than an impersonation target

Voice acting is not merely a set of phonetic sounds. Performers make choices about cadence, emphasis, timing, texture, emotion and character. A familiar character voice can carry years of audience recognition, but its value also comes from the ongoing labor and interpretation of the person behind it.

Related coverage includes Masako Nozawa Condemns Unauthorized AI Voice Use as Japan Actors Union Sets New Guidance.

Nozawa’s comment about an AI voice lacking heart is an artistic judgment, not a technical claim that a model cannot generate natural-sounding audio. Its importance is that the standard at issue is bigger than resemblance. A synthetic replica might sound superficially close and still not possess the performer’s intention, judgment or authorship. For actors, the concern is therefore both personal and economic: an unapproved model may exploit a recognizable professional identity while creating material that the performer did not choose or control.

That is especially relevant for games, animation, social media and other media built around recognizable vocal performances. A voice can be reused in dialogue, parody, advertising or posts that appear to carry an actor’s endorsement. The union’s approach identifies the actor, not the technology provider alone, as the person whose approval needs to anchor the process.

The guidance does not amount to a blanket statement that AI-related voice work can never happen. Its terms leave room for licensed use, provided the actor gives permission and is paid. Some Japanese voice actors have likewise supported systems that enable private, licensed voice replication.

That is an important practical division. “AI” is often used as a catch-all word, but the core dispute here is not simply whether software is involved. It is whether a performer can decide how their voice is collected, trained upon and used—and whether they receive compensation if that use moves forward.

By classifying authorized generated voice as a performance, the union’s guidance seeks to avoid a framework in which a company pays only for an initial recording session and then treats later synthetic output as cost-free. Salaries and royalties are different forms of compensation, but both recognize that subsequent use can have value. The supplied guidance does not set payment rates or describe a required contract structure, so those details remain outside what has been established publicly.

A developing response from Japanese voice performers

The union’s proposal arrives after years of visible concern from performers in Japan. In 2024, voice actors formed an organization focused on opposing AI voice misuse, citing the absence of legal copyright protections as a central issue.

Kenjiro Tsuda, whose roles include work in Attack on Titan, Chainsaw Man and Yu-Gi-Oh!, previously brought a case involving videos on TikTok that used his voice. The case was dismissed after the videos were removed. Even so, the incident illustrates the speed problem surrounding social platforms: disputed clips can circulate quickly, while remedies may depend on removal after the fact.

The Japan Actors Union’s guidelines are among the first in Japan focused on AI voice recordings. Guidance is not the same thing as a new law, nor does the information available establish a particular enforcement mechanism or penalty. But a collective standard can still clarify what performers consider acceptable conduct and what they expect from companies that want to build or use voice-replication systems.

What this could mean for game and media companies

For studios, publishers and AI firms, a consent-first approach implies that voice records cannot be treated as endlessly reusable by default. The safest reading of the union’s principles is straightforward: obtain permission before an actor’s voice is used for training, obtain permission for authorized deployment, and compensate the actor when the generated performance is used.

That does not answer every future question. The guidance does not, on the facts available here, resolve how consent would be documented, how long it would last, whether it could be withdrawn, or how a model trained on numerous performers would be audited. Those are the implementation questions that contracts, industry practice and potentially government policy will have to confront.

Still, the framework gives performers a clear argument against the idea that a voice model is separate from the individual whose work enabled it. It also gives companies an incentive to discuss the boundaries of use up front rather than assuming that a recording made for one project can become training data for others.

The issue sits within a broader Japanese policy debate over generative AI, including copyright infringement and cyberattacks. The government is facing questions about how to respond to those challenges. The union’s intervention adds a labor and performer-rights perspective to that conversation, centered on a basic proposition: a recognizable voice is not free for the taking because it can be digitized.

For the entertainment business, that point may prove more durable than the immediate dispute around any one AI clip. Fans often talk about a character voice as if it were a fixed attribute of the character. Nozawa’s remarks underline the human reality behind that shorthand. A voice performance comes from a person, and the union’s guidance argues that the person must retain a meaningful say in its AI future.

Concerns about technology’s effect on creative work are also being felt across game development communities, including places celebrated for studio-building expertise such as Skövde’s game-development scene. The specific rules and legal context may vary, but the question remains consistent: when technology scales creative output, who has the right to approve and benefit from the work that made that scale possible?