Various AI generated songs by “AI Sun Yanzi” hit the screen the other day. Related intellectual property issues became hot topics on media. I would like to discuss legal issues relating to songs generated from real person’s voice extracted by AI engines.

In brief, based on intellectual property law, it is not very likely that songs generated by AI trained with real person’s voice and used for non-commercial purpose infringe other person’s rights, but before they are made available online, licenses for copyright, backing music and elements used in videos should be obtained. In Civil Code, however, songs generated by AI trained with real person’s voice must be subject to the person’s authorization or otherwise, could infringe personal rights. Let’s look at the “AI Sun Yanzi” case.

Question 1: Is the intellectual property license required for training AI with Sun Yanzi’s voice?

Not exactly. Here is an interesting legal issue. People should have intellectual property owner’s license to train AI with their written or graphic works by using a word generation engine like ChatGPT or picture generation engine like Midjourney. Without such a license, they might be alleged to commit infringement. However, this is not correct in the case of training “AI Sun Yanzi” with her voice.

The “AI Sun Yanzi” exists on the condition that AI can obtain data containing special characteristics of her voice. According to how AI works, her voice is gathered from her speeches, not her albums. There are many files containing her voice, for example, ones made at interviews, conferences or concerts. You can obtain data containing special characteristics of her voice by training AI with these audio or visual recordings put into AI pronunciation acquisition engines.

Files containing her voice mean audio or video products in law. People making such products have the audio recording or video maker’s right. In addition, according to Copyright Law, as the performer in the audio and video products, Sun Yanzi also has the performer’s right in the content of the sound files. However, the audio recording and video maker’s and the performer’s rights are neighboring rights in Copyright Law that only protect traditional acts of reproduction, publication, transmission through information networks, etc., without exhaustive protection. Using such rights in a way that is not specified in law could not be deemed as infringement.

The way that AI engines use works by training AI with sound files put into AI engines did not exist when the Copyright Law was made. This involves no infringement of the audio recording or video maker’s and the performer’s rights.

Words and pictures used by ChatCPT and Midjourney in AI training are given strong protection by Copyright Law that includes “other rights of the copyright owner” in addition to specified ones. The act of training AI engines with word and graphic works protected by copyright, which even had not existed before the law was made, might infringe “other rights that the copyright owner should have”.

Question 2: Which intellectual property licenses are needed for publishing songs sung by “AI Sun Yanzi” online?

The publisher of White Hair, the most popular song on Bilibili created by Jay Zhou sung by “AI Sun Yanzi” should at least have the following intellectual property licenses.

1. Copyrights in lyrics and music of the song.

2. A license for the original audio recording of the backing music in the song. 

3. Copyrights in photos of Sun Yanzi in the video.

The White Hair sung by “AI Sun Yanzi” without any of the above licenses would involve intellectual property infringement risk. If it is used by the publisher personally, is it the fair use, a reason for exemption of liabilities in Copyright Law? No. The personal use of AI generated works means nobody but the publisher listens to the songs. Publishing them online is an act of transmitting information through networks which should be done with authorization.

Another question: if the song is produced by the publisher using AI to replace the original voice in the White Hair by Jay Zhou with the voice of Sun Yanzi, should the publisher have licenses for the audio recording maker’s right and the performer’s right of Jay Zhou relating to the original voice in the White Hair by Jay Zhou? It was reported that an expert believed that “AI imitation, publication or display of a singer’s voice is an infringement of the singer’s performer’s right”.

In my personal opinion, as stated above, the performer’s and the audio recording maker’s are neighboring rights, so with no restrictions in Copyright Law, use of AI to replace original voice with other person’s voice in audio or video products should not be found as infringement. The use of the original backing music in the White Hair sung by “AI Sun Yanzi” without authorization might be found as infringement.

The above parts are limited to current acts of “AI Sun Yanzi”, i.e., acts of internet users using AI for publishing and producing songs like ones sung by Sun Yanzi on their own, not involving the use of such songs for commercial purposes such as sale of albums and advertisement. A commercial use of Sun Yanzi’s voice to train AI, produce songs, etc. could violate the Unfair Competition Law and be found as unfair competition acts against generally accepted business ethics. If Sun Yanzi seek protection of her rights, the producer and the publisher might face a large amount of compensation.

Question 3: Does “AI Sun Yanzi” infringe Sun Yanzi’s personal rights?

In Civil Code, a civil entity’s personal rights are protected by law and cannot be harmed by any organization or person. Names, portraits and voice are part of the personal rights. The online publication of songs by “AI Sun Yanzi” in the name of Sun Yanzi involves personal right protection issues.

In Civil Code, no person can infringe other’s name by misuse or falsification. I don’t think that “AI Sun Yanzi” used her name to publish songs generated by AI using her voice in a way that infringed her name by misuse or falsification.

In Civil Code, no person can make, use or publish a person’s portrait without consent of the person.

“AI Sun Yanzi” used her portrait in the published video of White Hair in a way that may be deemed as portrait infringement. In Civil Code, fair use means using a person’s published portrait in study, art, classroom or scientific research on a need-to-use basis. However, the act of publishing songs online is not done for personal study or artistic purpose and does not fall in the scope of fair use.

To protect civil entity’s voice, Civil Code provides that no person can make, use, publish a person’s sound without consent of the person. The publisher of “AI Sun Yanzi” using AI technology to imitate her voice may have committed infringement by making, using or publishing the voice without permission.

Our team discussed whether “AI Sun Yanzi” infringed other person’s voice. Some believed no. Before the advent of AI, when some people imitated other person’s voice, sometimes on TV, nobody felt they were wrong. Why AI can’t imitate human voice as a person does?

I think human and machine imitations are different. People able to imitate other person’s voice are talented and well trained and few. Voices produced by human imitation may not be similar to real ones or exist in large amounts, so it’s innocuous. While voices imitated by AI directly using data extracted from voices are very similar to real one. The threshold is so low that average people with a little knowledge about computers following the AI training instructions accessible online can easily learn AI imitation training. AI imitations may exist in large quantities. AI abuse is terrible. Therefore, I think it necessary to require licenses for AI imitations.

Finally, intellectual property issues relating to artificial intelligence are not serious within the current legal framework. Artificial intelligence is emerging and generally accepted. Sun Yanzi has not taken any legal action against “AI Sun Yanzi”. Actually, ethical and other issues arise from the use of AI to imitate the ways that people write, take photos, speak, and make videos. We should take the good as well as prevent the bad effect of technological advances.