Facts
The plaintiff, a voice actor surnamed Yin, found that works voiced with the plaintiff’s voice were circulating widely on several well-known apps. Tracing showed that the voice came from a text-to-speech product offered on a platform run by a Beijing technology company, the first defendant. The plaintiff had earlier recorded audiobooks for a cultural media company, the second defendant, which held copyright in the recordings and provided them to a software company, the third defendant. The software company used the recordings as material to develop an AI voice product. The fourth defendant, a Shanghai network technology company, operated the cloud service platform on which the product was sold; the fifth, a Beijing technology development company, sold the product to the first defendant under an online service sales contract.
Question
The court had to decide whether a natural person’s rights in their voice, protected as a personality interest, extend to an AI voice synthesized from their recordings, and on what test a synthetic voice counts as identifiable. It also had to decide whether the media company’s copyright in the recordings entitled it to authorize the use of the plaintiff’s voice for AI development and a commercial text-to-speech product. Finally, it had to determine which of the five companies in the chain were at fault and how liability should be allocated among them.
Decision
The court held that protection of a natural person’s voice extends to AI-generated voices, provided the voice is identifiable. Identifiability depends on whether ordinary listeners in the relevant field can associate the synthetic voice with the person, and here the AI voice was highly consistent with the plaintiff’s in timbre, intonation and style. Copyright in the recordings did not include a right to authorize AI processing of the plaintiff’s voice. The first and third defendants were ordered to apologize in writing, the second and third defendants were held jointly liable for RMB 250,000 in economic losses, and the remaining claims were dismissed.
Why it matters
This was the first Chinese judgment on personality rights in an AI-generated voice. It separates the copyright in a recording from the personality interest in the voice it captures, so a license to use recordings does not by itself authorize voice cloning. The identifiability test gives courts a workable standard for synthetic voices and is relevant to voice deepfakes more generally.
Related stages
Neither party appealed after the judgment was pronounced, and it has taken legal effect.