In April 2024 the Beijing Internet Court delivered a first-instance judgment in China’s first personality-rights case involving an AI-generated voice. The voice of a dubbing artist, Yin, had been used after AI processing. The court held that, where recognisability is established, protection of a natural person’s voice rights can extend to AI-generated voices, applying the test of whether ordinary listeners in the relevant field can identify the person. It ordered a written apology and damages of RMB 250,000.
The key holding
The case did not stop at whether the voice had been copied. The court anchored the test in recognisability: can ordinary listeners in the relevant field connect the AI-generated voice to a specific person. If they can, the voice falls within the protected scope; if they cannot, it is closer to a generic synthetic voice.
That standard bears directly on product design. Whether a voice-cloning feature infringes does not turn on technical parameters but on whom the output points to in the listener’s ear. Product teams should test “does this sound like a particular person”, not only audio quality and naturalness.
Sound recording copyright is not AI authorisation
The second important conclusion is that copyright in a sound recording does not equal permission for AI use. A licence to use a recording typically covers copying, distribution and making available online; it does not automatically cover AI synthesis of the voice or model training on it.
That conclusion changes how licensing chains are drafted. An older dubbing contract might say only that the recording may be used for a given project. It now needs separate treatment of AI synthesis, model training and derivative voices.
What the licensing chain should contain
A dubbing or voice licence should state at least: whether AI synthesis or voice cloning is permitted; the permitted purposes, channels and term; whether model training is allowed; how synthetic models and assets are handled after termination; and whether sub-licensing is permitted.
Companies using third-party voice libraries also need to trace upstream: did the library provider obtain those permissions from the rights holder, and does the scope cover how the company actually uses the voice. A licence to use a voice library is not a licence to use the voice through AI.
AI dubbing products and platforms
Products offering voice cloning or AI dubbing should prohibit unauthorised synthesis of another person’s voice in their terms and maintain complaint and takedown mechanisms. Platform liability depends on its control over content, but the speed of response after notice often shapes the finding.
Platforms can also act at the design stage: block voice samples clearly pointing to well-known figures, and verify provenance for high-frequency cloning requests. These measures do not eliminate infringement, but they lower the platform’s own exposure.
Allocating risk
Contracts can include rights warranties and indemnities requiring the voice provider to guarantee complete authorisation and to cover claims arising from defects. Such clauses do not remove infringement risk, but they shift loss to the party best placed to verify the licence.
For companies already operating AI dubbing features, a stocktake is the sensible first step: where the voice assets came from, what uses the licences cover, and whether the agreements mention AI synthesis. The results determine remediation priority.
For personality-rights and infringement response, see our AI infringement service; for voice asset rights and licensing, see IP protection; for licensing and indemnity drafting, see commercial contracts.
This article is based on the first-instance judgment in the AI-generated voice personality-rights case delivered by the Beijing Internet Court in April 2024, and on public information. It is for reference only and does not constitute legal advice.