Where Is the Line When AI Imitates Someone's Voice? Reading the Google Case for the Logic of Voice Protection

A voice is a natural person's unique identifier, bound up with individual dignity and carrying increasing commercial value. As AI speech synthesis and voice cloning proliferate, imitating another's voice commercially without permission has become a frequent source of legal dispute.

In January 2026, veteran NPR host David Greene sued Google in the Superior Court of Santa Clara County, California, alleging that Google's AI product NotebookLM — which includes a podcast-style feature — imitated his recognizable vocal characteristics, including his cadence, tone, and verbal mannerisms such as his use of "uh," in a commercialized product feature. The case drew global attention to AI voice infringement. It has since been removed to federal court on a jurisdictional dispute and remains in discovery.

The more widespread AI speech becomes, the more concentrated the legal questions. Does training AI on someone's audio require authorization? At what point does an AI voice infringe? How is commercial use defined? This note draws on U.S. and Chinese cases to set out how AI voice infringement is actually assessed.

I. The Core Dispute: Does Training AI on a Voice Without Permission Infringe?

AI speech models require large volumes of audio for training, extracting timbre, prosody, and intonation to synthesize new speech. That raises a key question: does using another's public audio to train a model without consent, and commercializing a similar output voice, infringe voice-related rights?

Technically, AI voice imitation does not necessarily require complete reproduction of the original recording; it extracts vocal features algorithmically and reconstructs them. But such non-reproductive use may still constitute "use" in a legal sense.

In Lehrman v. Lovo, in the Southern District of New York, an AI company obtained voice actors' recordings under the guise of "research testing" and used them without permission to train models and sell cloned voices commercially. While the court did not sustain the copyright claims, it recognized that commercial exploitation of personal identity indicia is actionable.

By comparison, in Chinese practice, the Civil Code expressly provides that protection of a natural person's voice applies the rules governing portrait rights by reference. The Yin v. [AI Technology Company] decision held that voice personality interests are independent: because training material requires use of the natural person's original voice, using and processing that voice through AI without permission infringes voice rights.

II. The Key Element: The Voice Must Be Identifiable

In both the United States and China, identifiability is the foundational element — an ordinary listener must be able to associate the AI voice with a specific person, without the match needing to be exact.

At the motion to dismiss stage in Lehrman v. Lovo, the plaintiffs alleged that the cloned AI voice was nearly identical to their own and highly identifiable. The court found that sufficient to survive dismissal at the pleading stage and allowed the misappropriation of voice claim under New York law to proceed.

In Greene, the plaintiff likewise alleged that NotebookLM's AI voice closely resembled his own in timbre, intonation, and verbal habits, such that friends, family, and listeners could be confused — establishing identifiability.

In Chinese practice, beyond objective voiceprint identification and confirmation, courts also consider the manner of use, comparing the new voice's timbre, intonation, and delivery style against the natural person's voice, and applying a subjective standard based on the general public or a defined segment of the public in light of the person's public profile.

The AI infringement case involving the well-known actress Wen Zhengrong is a recent and representative instance of combined voice and likeness infringement. Unauthorized actors used deep synthesis technology to fabricate her facial image and vocal characteristics for livestream commerce and promotional purposes, in content sufficient to confuse ordinary members of the public and therefore highly identifiable, seriously infringing both her voice and portrait rights. The matter has been opened as a criminal case — indicating not only the civil character of AI-forged identity indicia but that, where the circumstances are serious, criminal liability may attach.

III. Commercial Use: Any Profit-Making Purpose May Fall Within Scope

California Civil Code § 3344(a) expressly brings voice within the right of publicity: using another's voice without consent for products, advertising, sales, or solicitation constitutes infringement.

"Any person who knowingly uses another's name, voice, signature, photograph, or likeness, in any manner, on or in products, merchandise, or goods, or for purposes of advertising or selling, or soliciting purchases of, products, merchandise, goods, or services, without that person's prior consent…"

In Greene, the AI voice at issue was a component of the NotebookLM product feature, through which Google enhanced user experience and product competitiveness — plainly commercial in character.

New York Civil Rights Law § 50 prohibits using a living person's name, portrait, picture, likeness, or voice for advertising or trade purposes without written consent:

"A person, firm or corporation that uses for advertising purposes, or for the purposes of trade, the name, portrait, picture, likeness, or voice of any living person without having first obtained the written consent of such person… is guilty of a misdemeanor."

In Lehrman v. Lovo, the defendant packaged and sold cloned AI voices on a subscription model — a classic commercial exploitation — and the court found the predicate for a right of publicity claim satisfied.

U.S. law emphasizes the proprietary protection of state right of publicity; China protects voice interests within its personality rights framework and defines commercial use more broadly. But in both systems, an AI voice infringement finding turns on three points:

• use without the rights holder's permission;
• an identifiable AI voice; and
• a commercial or profit-making purpose.

IV. Why AI Voice Cases Are Usually Not Copyright Cases

AI voice disputes appear to originate in audio scraping and model training and are often mistaken for copyright matters. In practice, plaintiffs rarely proceed on copyright, asserting instead voice personality interests, the right of publicity, or voice rights.

The underlying reason is that copyright protects a specific expression fixed in a medium, not the identity-identifying value of a voice itself. A recording may be a copyrightable work, but a natural person's distinctive timbre, intonation, and verbal habits — the vocal identity features that identify a particular individual — derive their core value from personality identification and commercialized personality interests, not from the fixed expression copyright protects.

In Greene, the complaint pleaded no copyright claim at all, proceeding instead on California statutory and common law right of publicity, unfair competition, and unjust enrichment. The core theory was not that Google had copied a copyrighted recording, but that Google had used his identifiable voice, expressive style, and personality characteristics for AI training and commercial output without permission.

In Lehrman v. Lovo, the copyright claim was dismissed. The court held that under 17 U.S.C. § 114(b), a sound recording copyright protects only the specific expression fixed in the original recording and does not extend to a new sound independently generated by AI; even where the AI simulates the original, absent direct copying of the recording there is generally no copyright infringement:

"The exclusive rights of the owner of copyright in a sound recording … do not extend to the making or duplication of another sound recording that consists entirely of an independent fixation of other sounds, even though such sounds imitate or simulate those in the copyrighted sound recording."

That reasoning is precisely the point: the essence of an AI voice dispute is not which recording was copied but whose vocal identity was appropriated. AI may not copy the original recording at all, yet may still take another's identifiable vocal features, personality indicia, and commercial appeal — causing both personality harm and market substitution.

This article addresses general legal questions only and does not constitute legal advice on any specific matter.