Court ruling in China: AI-generated voice imitation violates personal rights

updated on 3 September 2024 | reading time approx. 2 minutes

In April 2024, the Beijing Internet Court issued another landmark ruling on the use of artificial intelligence (AI), after the same court ruled in November 2023 that AI-generated images can be copyrightable. The new ruling concerns the infringement of personal rights by AI-generated voice imitations.

Background of the case

The plaintiff in the case, a voice artist, discovered that her voice had been used in numerous audio books without her consent. Through her own research, she discovered that audio recordings of her that she had previously made for audio books with a media company had been licensed by the company to an AI software company with permission to commercially use, reproduce and modify the audio recordings. This company used the voice to train an AI-based tool (text-to-speech application) and marketed the application with the plaintiff's voice. Another company acquired the application and used it to produce various audio books.

The court found that the defendants had used the plaintiff's voice without her consent, thereby violating her personal rights, and ordered the defendants to pay damages and apologize.

Specificity of the case

A key question for the court was whether a person's right to their voice includes the voice reproduced by the use of AI. The court's answer was in the affirmative. A person's natural voice can be identified by its tone and frequency. It has unique and consistent characteristics that are capable of evoking thoughts and feelings associated with that person. Minor changes to the voice through the use of the AI application do not prevent listeners from associating the voice generated by the application with natural voices, thereby evoking the thoughts and feelings associated with the original person and directly associating the voice heard with that person. This reveals the person's identity.

In the proceedings, the court found that the AI-generated voice and the plaintiff's voice were very similar in timbre, tonality and pronunciation style, which could lead listeners to associate or attribute the AI-generated voice to the plaintiff herself. The AI-generated voice is identifiable as the plaintiff's voice, so that the plaintiff's rights in her voice can be extended to the AI-generated voice.

According to the Chinese Civil Code, the provisions on the protection of the right to one's own image also apply to the protection of the right to one's own voice. Accordingly, any natural or legal person is prohibited from using, falsifying or manipulating a person's voice using technology. The law prohibits the production, use or distribution of a person's voice without his or her consent.

Conclusion

The decision of the Beijing Internet Court is likely to set the standard for all cases in which an AI-generated voice can be traced back to the voice of a natural person. If the person has not consented to the use of his or her voice, the person's personal rights have been violated, which may lead to a claim for damages.

However, even if there is a contract with a person, e.g. for the production of an advertisement, the recordings cannot simply be used for other purposes. According to the Beijing Internet Court, the purpose and scope of a person's original contractual consent to use, for example, their voice, must be considered.

The training of generative AI requires the use of a variety of existing data, such as text, images, videos, etc. Many of these sources may be copyrighted. The use of such sources for training may therefore involve a risk of copyright infringement.