With artificial intelligence technology advancing at a rapid pace, applications such as speech synthesis and voice cloning have gradually become part of everyday life. Yet convenience brings hidden concerns: when someone collects another person's voice without permission to train AI, synthesizes speech from it, and profits commercially, how should legal liability be defined?
According to the WeChat public account of the Shanghai High People's Court, on September 29, 2026, the Shanghai First Intermediate People's Court concluded Shanghai's first dispute over the protection of a natural person's voice rights arising from AI-synthesized speech, ruling that the platform operator had committed voice infringement and ordering it to pay the rights holder 50,000 yuan in economic damages.
Ms. Wang is a voice actor verified under her real name on a certain app, where she regularly performs dubbing and publishes works to earn income. In January 2025, Ms. Wang learned from a friend that Company A had apparently used her voice in an internet promotion and user acquisition campaign. After verification, Ms. Wang believed the voice used in the campaign was likely AI-synthesized, so she notarized the relevant audio to preserve evidence and filed a lawsuit with the people's court.
Ms. Wang argued that she had never cooperated with Company A or granted it any authorization, and that Company A's unauthorized use of her voice constituted infringement. She demanded that Company A stop the infringement and pay 300,000 yuan in economic damages. Company A argued that the audio in question was indeed generated by AI it had developed, but that the specific source and the training materials fed into it could not be confirmed because the former employee involved had left the company. The company said it had previously been unaware of Ms. Wang's identity as a voice actor, that it had never collected or used her voice, and that the existing evidence could not directly prove that her voice was used in its promotional campaign.
Image sourced from the internet. During the trial, to prove the facts of the infringement, Ms. Wang conducted a forensic analysis comparing the notarized audio in question with her own voice recorded on the spot. The appraisal opinion showed that among 28 formant acoustic indicators, 24 of the audio in question deviated from Ms. Wang's own audio by less than 10%, with 16 of them below 5.36%, and the portions that were relatively similar and highly similar reached 90%.
The court of first instance held that Company A's conduct constituted AI-generated voice infringement and caused economic loss to Ms. Wang, and accordingly ordered Company A to pay her 50,000 yuan in compensation. Both parties were dissatisfied with the first-instance judgment and appealed.
The Shanghai First Intermediate People's Court held after trial that using a natural person's voice as training corpus without that person's consent to imitate the person's timbre, intonation and pronunciation style, and thereby generate a synthesized voice capable of identifying that person, should be deemed an infringement of the natural person's voice rights. In this case, Ms. Wang, as a voice actor, had presented evidence proving that her dubbing works were published on public online platforms, and that Company A had information channels and the possibility of obtaining the source of her voice. At the same time, the appraisal report showed that the synthesized voice deviated only slightly from the human voice audio and that the two were highly similar, enough to lead ordinary members of the public to associate them with a specific identity, giving the voice identifiable characteristics. Although Company A denied this, as the party controlling the original training data it failed to fully prove the matters to be proven, such as the lawful source of the training materials, that it had not used the natural person's voice characteristics, and that it had not caused confusion or misidentification among the public, and it should bear the adverse consequences of failing to discharge its burden of proof. On this basis, it could be found that Company A, without Ms. Wang's consent, commercially used a synthesized voice capable of identifying an individual's identity toward the public, which constituted voice infringement.
At the same time, regarding the determination of the amount of damages for voice infringement, the standard of licensing fees for similar uses by the rights holder should be given priority as a reference; if the rights holder's evidence is insufficient and the licensing fee is difficult to ascertain, the amount of loss may be comprehensively determined by taking into account factors such as the duration of the infringement, the scope of dissemination, the manner of use, the degree of fault, the result of the damage, and the proportion of contribution of personality elements to commercial exploitation. In this case, the voice at issue was online for a relatively short period and was a procedural broadcast prompt sound effect, rather than the core selling point or traffic-driving vehicle of the campaign. In addition, Ms. Wang's online following and public activity were not high, and the standalone commercial market value of her voice was limited. Therefore, the court held that the first-instance court's discretionary award of 50,000 yuan in damages was fair and reasonable.
In summary, the Shanghai First Intermediate People's Court ruled in accordance with the law: the appeal is dismissed and the original judgment is upheld.