China’s First Case regarding AI-generated Voice Infringement

I. Introduction

On April 23rd, 2024, the Beijing Internet Court ruled on China’s first case regarding infringement of personal rights (known in the U.S. as “publicity right”) involving AI-generated voices.[1] In its decision, the Beijing Internet Court found that the defendants used the plaintiff’s voice in their AI text-to-speech applications without authorization and thus violating the plaintiff’s personal rights related to voices.

II. Background

The plaintiff, Ms. Yin (the “Plaintiff”), is a voice actor. Plaintiff noticed a few dubbing works that used her voices were widely circulated on several popular applications. After voice filtering and source tracing, she discovered that the voices in those dubbing works came from an AI text-to-speech application (the “TTS AI Application”) operated by a Beijing AI technology company (“Defendant A”), which services allow users to turn prompts into speeches.

Plaintiff had a voice recording contract with a media company (“Defendant B”), who owns the copyright of the sound recordings. It turns out that Defendant B then licensed Plaintiff’s sound recordings to an AI software company (“Defendant C”), allowing Defendant C to use, copy, and modify the data in the sound recordings for commercial or non-commercial purposes. Defendant C then used Plaintiff’s sounds as training data to train the TTS AI Application, which application was later adopted by Defendant A through interfacing with Defendant C’s applications, as hosted by a cloud server company (“Defendant D”) and through a distributor (“Defendant E”).

Plaintiff claimed that the defendants' actions infringed upon her voice rights, and that Defendants A and C should immediately cease the infringement and make an apology. Plaintiff also claimed that all defendants should compensate for her economic losses and mental distress.

Defendant A, which adopted the TTS AI Application, argued that as a bona fide third party, it legally purchased the TTS AI Application through regular distribution channel and had fulfilled its reasonable duty of care, and thus shall not be held liable for infringement.

Defendant B, who owns the recording contract with Plaintiff, argued that it owns the rights and permission to license the sound recordings, including Plaintiff’s voices, to Defendant C.

Defendant C, the developer of TTS AI Application, argued that it had obtained the necessary authorization for the voices in dispute and the final voices generated through TTS AI Application are distinguishable from Plaintiff's voices.

III. Issues

In this case, the court focused on the following issues: (1) whether Plaintiff's voice rights can be extended to the AI-generated voice in dispute; (2) whether Defendant B and Defendant C have obtained the necessary legal authorization to use Plaintiff’s voices for AI training purposes; and (3) if the defendants were found liable for infringement, what civil liabilities should each of them bear.

1. Whether Plaintiff's voice rights can be extended to the AI-generated voice in dispute.

The court started its analysis by quoting Article 1023 of the Chinese Civil Code, which stipulates that the protection of a natural person's voice should be treated the same as the protection of his image right,thus confirming the protection for voice rights as a special personal right.[2]

With regard to the issue whether the voice right of a natural person could be extended to AI-generated voices, the court acknowledged that a natural person’s voice can be distinguishable by its voiceprint, timbre and frequency, and thus has unique, singular, and stable characteristics. Human voices can cause others to form or induce thoughts or emotional activities associated with that natural person and can also reveal that natural person's behavior and identity to the outside world. The voice of a natural person can become identifiable after repetition or long-term listening by others. Therefore, the court ruled that, if an artificially synthesized voice can cause the general public or the public in the relevant field to associate the voice with that particular natural person based on its timbre, tone, and pronunciation style, then the voice should be deemed identifiable.

In this case, because the AI-generated voices are highly consistent with Plaintiff's own voice timbre, tone, and pronunciation styles and can cause the public to relate or attribute the AI voices with Plaintiff herself, the court held that the voices are identifiable and that Plaintiff's voice rights can be extended to the AI-generated voice in dispute.

2. Whether Defendants B and C obtained the necessary authorization from Plaintiff to use her voices for AI purposes.

The court held that, in this case, Defendant B owns copyright to the sound recordings, but does not own the rights to authorize others to use Plaintiff's voices in the sound recordings for AI training purposes. Therefore, the court ruled that Defendant B and Defendant C failed to obtain the necessary authorization to use Plaintiff’s voices for the AI training purpose.

3. If the defendants were found liable for infringement, what civil liabilities should each of them bear.

The court held that the unauthorized use of Plaintiff's voices constitutes infringement and factors such as the circumstances of the defendants’ infringement and the value of similar market products should be taken into consideration when evaluating the damages.

In this case, the court held that Defendant B and Defendant C used Plaintiff's voices for AI purposes without her permission, thus finding them liable for infringing Plaintiff’s voice rights. With regard to the other co-defendants (Defendants A, D and E), the court ruled that they did not have subjective fault and thus shall not bear the responsibility for damages.

In summary, the court ordered that Defendant A and Defendant C to apologize to Plaintiff, and Defendant B and Defendant C to compensate Plaintiff for a total loss of RMB250,000. The court rejected Plaintiff’s other claims.

IV. Comments

This is China’s first case that expressly protects against the use of AI to infringe on an individual’s voice rights. We found several interesting similarities between this Chinese case and the new Tennessee law Ensuring Likeness, Voice and Image Security (ELVIS) Act that was passed last month in March 2024. First, both the Chinese existing law and the ELVIS Act recognize the protection of an individual’s voice rights, in addition to the rights of an individual’s name, image and likeness. Second, both the Beijing court and the ELVIS Act clearly protect against the use of AI to infringe on someone’s voice rights, where the Beijing Internet court found the AIGC developer liable for its unauthorized use of Plaintiff's voices for AI training purposes. The convergence between this case and the legislative developments at other jurisdictions shows a shared commitment to protect the publicity rights, including individual’s voices, in the age of AI.

But, we also noted a sharp “contrast” between the ELVIS Act and this Chinese decision. In the ELVIS Act, those recording companies, who have an exclusive contract with the recording artist or an exclusive license to distribute sound recordings featuring the individual’s voices, are entitled to bring an action on behalf of their recording artist to enforce the relevant publicity rights. However, in this Chinese decision, the Beijing Internet Court clearly distinguished the copyright/license in a sound recording from a license/permission to use the recording artist’s voices for a different purpose other than the sound recording, and therefore found the recording company (Defendant B) liable for infringement. It would be interesting to watch the development of contract negotiations and drafting between label/media companies and their recording artists and talents in the future.