The Brief
The Shanghai Intellectual Property Administration has issued an intellectual property registration certificate to Bilibili for its "Luo Tianyi Audio Dataset," marking the city's first formal data-product IP registration for a fully virtual singer. Granted under Shanghai's interim data registration regulations, the certificate recognizes the dataset as an original commercial product resulting from substantial processing and innovative labor. The move aims to curb unauthorized generative AI voice cloning by establishing upstream rights and providing preliminary judicial evidence for future infringement claims.
Why it matters
The rapid spread of generative AI voice synthesis has made vocal timbres easy to scrape, train on, and clone with minimal cost. Because traditional copyright frameworks historically struggled to define ownership over synthetic vocal timbres detached from specific musical compositions, Shanghai's issuance of an IP certificate for a curated virtual singer dataset creates an operational precedent. Establishing "substantial processing" and "innovative labor" as benchmarks allows tech companies to establish prima facie evidence of data property ownership, lowering litigation and enforcement costs against unauthorized AI training.
China context
The certification was processed under the Shanghai Interim Measures for Intellectual Property Registration and Deposition of Data Products, reflecting local implementation of China's broader national framework to treat data as a factor of production. Shanghai has taken an active role in testing intellectual property and assetization mechanisms for digital assets, seeking to resolve regulatory and legal ambiguities where rapid AI deployment outpaces legacy statutes.
Editor's View
EDITOR'S VIEW — Analysis and inference, not factual reporting.
Shanghai's initiative to certify the Luo Tianyi dataset represents a pragmatic shift toward administrative registration to solve a vexing generative AI problem. While traditional copyright protects completed sound recordings and lyrics, synthetic personas rely on underlying corpus tuning and acoustic parameters that unauthorized models can mimic without directly re-broadcasting copyrighted songs. By shifting the verification step upstream to the dataset itself, authorities are granting virtual human operators an actionable legal shield, though the real test will lie in how effectively courts outside Shanghai uphold these certificates in cross-jurisdictional AI scraping disputes.
What to watch
- Bilibili's planned follow-up application to register a more granular 'speech corpus unit library' for Luo Tianyi.
- Judicial rulings in Chinese courts testing whether the data product registration certificate holds up as conclusive evidence in AI voice-cloning disputes.
- Whether other virtual idol management agencies and AI voice studios across China follow suit by registering their proprietary synthetic training corpuses.
Key Takeaways
- 1The Shanghai Intellectual Property Administration issued an IP registration certificate to Bilibili for the 'Luo Tianyi Audio Dataset.'
- 2Officials determined the dataset met legal benchmarks for 'substantial processing' and 'innovative labor' under Shanghai's interim data product rules.
- 3The certificate serves as preliminary judicial evidence of lawful data ownership to aid enforcement against unauthorized generative AI voice cloning.
- 4Bilibili plans to seek additional IP registrations for fine-grained speech corpus libraries to explore tiered rights frameworks.
The Shanghai Intellectual Property Administration has granted an intellectual property registration certificate to Shanghai Bilibili Technology Co., Ltd. for its "Luo Tianyi Audio Dataset," state news agency Xinhua reported. The filing represents an intellectual property registration specifically targeted at an audio data product derived from a fully virtual human.
Luo Tianyi, launched in Shanghai in 2012, is a well-known Chinese virtual vocalist operated by Bilibili. Over more than a decade, the character has relied on voice synthesis technologies to generate singing audio, gradually extending her presence into commercial music composition, stage performances, and live-streaming interactions. According to officials, the registered audio dataset was developed through the systematic collection, annotation, and acoustic calibration of her distinctive vocal timbre.
Chen Huiqing, deputy director of Shanghai's Yangpu District Market Supervision Administration, stated that the dataset exhibits originality and commercial value. Chen noted that the product satisfies the statutory standards of "substantial processing" and "innovative labor" set out in the Shanghai Interim Measures for Intellectual Property Registration and Deposition of Data Products.
The regulatory milestone comes amid growing industry scrutiny over generative artificial intelligence, which has enabled rapid and unauthorized cloning of signature voice profiles. Officials pointed out that under existing judicial precedents, a data product intellectual property registration certificate can serve as preliminary evidence that a rights holder possesses legitimate property interests in data and that the underlying collection methods were lawful. Such documentation is intended to lower the burden of proof and litigation costs when enterprises confront infringement.
Chen added that the registration establishes a practical template for digital asset protection. By shifting the rights-confirmation process upstream, the mechanism aims to clarify ownership early, providing administrative and legal backing before disputes escalate.
Looking ahead, Bilibili plans to seek intellectual property registration for an even more granular "speech corpus unit library" tied to Luo Tianyi, according to Zhu Yao, the company's senior legal director. Zhu indicated the enterprise intends to further explore tiered paths for confirming rights over virtual human voice data.