The Pulse
Shanghai Court Upholds ¥50,000 Award in AI Voice-Cloning Case
Shanghai’s First Intermediate People’s Court upheld a ¥50,000 award after finding that a company used an identifiable AI-generated version of a voice actor’s voice without permission. The ruling says the company, which controlled the traini
AI.info Team · Written with AI ·
Shanghai’s First Intermediate People’s Court on September 29 rejected appeals by both sides in a voice-cloning dispute, leaving in place an order for a company to pay a voice actor 50,000 yuan. The court found that the company used an AI-generated voice identifiable as hers in an online promotion without her consent. The decision came as China’s Supreme People’s Court was setting out national guidance on AI disputes, including the use of a person’s voice as training material.
A promotion led to a voice comparison
The plaintiff, identified in the court account as Ms. Wang, is a voice actor who publishes work on an app for payment. In January 2025, a friend told her that a company’s online user-acquisition campaign appeared to use her voice. Wang had the audio notarized and sued in Shanghai’s Changning District People’s Court, seeking 300,000 yuan and an order to stop the use.
The company, identified only as A, said the audio was AI-generated but argued that it could not confirm what material a former employee had supplied to the system. It also said it had not known Wang’s identity and that the evidence did not directly establish that the promotional audio used her voice. Wang’s side commissioned a forensic comparison of the disputed recording and a recording made of her voice in person.
The examination found that 24 of 28 acoustic formant indicators had deviations below 10%; 16 were below 5.36%. The court account described the portions assessed as similar or highly similar as reaching 90%. Those findings supported the court’s view that ordinary listeners could associate the synthesized voice with a specific person.
The company held the training-data evidence
The appeals court said using a person’s voice as training material, then imitating that person’s timbre, tone and speaking style to generate an identifiable synthetic voice without consent, infringes voice rights. It also pointed to Wang’s publicly available voice work as a possible route by which the company could have obtained samples. The court said A controlled the original training data but did not adequately prove that the supplied material was lawful, that the output avoided her vocal features, or that it could not be mistaken for her voice.
That allocation of proof addresses a practical problem in disputes over model training: people outside a company generally cannot inspect its source recordings or internal logs. The court said a claimant may make an initial showing by establishing a reasonable way the voice could have been obtained and a high degree of similarity in the generated audio. The company then must produce evidence on the training material’s legality or show that the output is unrelated to the claimant’s voice.
“Technological neutrality does not mean neutrality in use.”
Wu Dan, judge in the Shanghai First Intermediate People’s Court’s filing division (translated from Chinese)
Wang and the company both appealed the first-instance judgment. The intermediate court dismissed both appeals and affirmed the 50,000-yuan award, along with the company’s obligation to cover necessary litigation expenses. The lower court had weighed the audio’s duration and reach, the harm to Wang and the company’s degree of fault in setting compensation.
China’s September guidance addresses cloned voices
The ruling followed an AI-disputes opinion issued by China’s Supreme People’s Court on September 7. Article 4 says that, absent a legal exception, using someone’s voice as training material to mimic their vocal qualities and generate an identifiable synthetic voice without consent can support a claim for infringement of voice rights. The national court’s guidance and the Shanghai decision address the same conduct, though the case account describes a dispute between Wang and one company, not a nationwide precedent designation.
The Shanghai court’s finding turned on two concrete points: the generated audio was sufficiently identifiable, and the company that controlled the training data did not substantiate its account of how that data was obtained or used. The final order requires A to pay Wang 50,000 yuan and her necessary litigation expenses.