Those who pay attention to AI may still remember the previous "AI Stefanie Sun". In fact, it only takes 10 minutes to 1 hour of audio material, plus a decent Nvida graphics card, and some time and electricity, everyone can create a similar "AI singer" sound model.

As in the past, technology giants share models with the concept of "only for learning" and educate "all beings", and resource communities are also gradually forming among the people, but in the entire AI world "vibrant" and increasingly flourishing At that time, a "copyright cannon" had already detonated in America...

The Discord community that was "banned"

There is a community (server) called AI Hub on Discord. Netizens upload the So-Vits-SVC/RVC sound model they have "alchemy" (trained) to here, making it a well-known AI speech synthesis over time. Model distribution center, there are already more than 140,000 Discord users active here.

Under the voice-model channel, we can see a wide variety of voice models. For the two-dimensional characters alone, there are Ram in "Re:0", Saber in "Fate", and Inosuke in "Demon Slayer: Blade". Other well-known figures:

You can also see singers from various countries such as Michael Jackson, Ayumi Hamasaki, and Aimer "competing on the same stage" across time and space:

You can also see Rappers come to RAP Battle in the AI ​​world, and the whole community has the feeling of "AI Quyuan Miscellaneous Forum":

However, recently, a letter from Discord broke this "AI feast". Discord stated that they had received a notice of infringement from the RIAA, and the other party wanted Discord to shut down the server of AI Hub and disclose the identity information of several users to it . Fortunately, Discord has not directly shut down the server as requested by the RIAA, but the administrators found that the "infringement" link mentioned in the letter is no longer accessible.

Although the scope of the spread is not large, the "sound ban" did happen quietly.

Who is the RIAA?

The RIAA, which launched the "copyright gun", is the Recording Industry Association of America , a trade group composed of a number of American record companies and representing the US record industry. In addition to formulating audio standards, music authorization, collecting royalties, and certifying sales, their main function is to combat infringement, sanction piracy and other litigation business.

Judging from the notification letter obtained by foreign media TorrentFreak, the main basis for RIAA’s request to shut down this time is “sharing copyrighted recording files”. Although TorrentFreak tried to ask for specific reasons, the RIAA indicated that it would be disclosed later. However, judging from the known information, training these "AI singer" sound models requires a certain period of original audio material, and according to the infringement evidence stated in the letter, some sharing posts in AI Hub with original training set files It may be the fuse of the "copyright gun" this time.

If shutting down the server is only about "things", then the DMCA subpoena is undoubtedly targeting "people". DMCA is generally called the "Digital Millennium Copyright Act of the United States" in China. We can often see overseas copyright owners picking it up in various copyright disputes. And this time, the RIAA obtained a DMCA subpoena in federal court in the District of Columbia, which means they can ask Discord to hand over information about alleged infringers .

In the letter accompanying the subpoena, they wrote "You must disclose to the RIAA information sufficient to identify the infringer. This will include the individual's name, physical address, IP address, phone number, email address, payment information, account updates and account history." Submission deadline is June 30.

Is the RIAA really "killing" AI? At least in response to foreign media, they emphasized that AI itself is not a problem, and the RIAA's enforcement efforts focus on those who exploit the works of artists without their consent. So will the offender go to jail? There's no sign of that, but the RIAA did say in court that it wants to learn more about the alleged infringers.

AI Hub's Response

AI Hub administrator .tea revealed in an interview with foreign media that although the RIAA warning letter is true, Discord has not yet officially notified them of the warning letter. And as administrators, they actually have their own responses to the RIAA's allegations.

Similar to what was said above, the administrators of AI Hub also believe that the source of the problem this time should be the original data set of infringing audio, and these infringing content is actually "zero tolerance" of AI Hub, and they don't mind deleting it. On June 24th and 26th, the AI ​​Hub administrator issued two documents, clearly informing users that sharing the "AI singer" voice model is not infringing, but sharing the original audio material is an infringement , and requires users to share the trained voice model* *Uploaded to the HuggingFace platform**, because HuggingFace can ensure that these sound models that only contain codes and voice pattern data will not be mistakenly removed due to "copyright guns".

However, compared to the calmness of the AI ​​Hub administrator, the netizens of the AI ​​Hub may react more intensely. In fact, the RIAA issued a similar rights protection action in late May, but it seemed to have the opposite effect at the time. At least the number of AI Hub users did not decrease but increased, from the original 80,000 users to 140,000 within a few weeks. After the "warning letter" incident broke out, some users even developed "reverse psychology".

"Reuploaded because I'm not afraid to go to jail ," wrote one user, reuploading Frank Sinatra's voice model. (It has been reported that Frank Sinatra's copyright holders don't like AI models, so some people think that this warning letter storm may have something to do with it.)

For the "AI Singer", there are other voices...

There have been various voices for these "AI singer" models.

The most famous "voice" in the Chinese-speaking circle is Stefanie Sun's blog post "My AI". In the article, she laughed and said, " At the moment, I feel like a person who is eating popcorn and sitting in the best seat in a movie theater. Everything is possible, and everything doesn't matter. I think pure thinking and being myself are enough. "

This is a singer's cheerful attitude towards mass entertainment. Perhaps it is because he has been famous for many years and has experienced ups and downs. On the contrary, he doesn't care much about the fact that the AI ​​character has become "the current top trend". Not to mention copyright issues, even for some fans' "picky" AI models, she also feels that technological development will overcome the shortcomings of these "AI singers".

In Japan, which pays attention to copyright, discussions on the copyright of AI voice models have also been put on the agenda. The Japanese Actors Union, which is committed to protecting the rights and interests of actors and voice actors, released the "Proposal on the Use of Generative Artificial Intelligence Technology" this month, calling for review of copyright . The operation of the law and the establishment of "sound likeness rights" .

Looking back at China, the CCTV news channel has also reported on the news that the "AI singer" may involve many legal issues , and interviewed professionals including professors from the University of Science and Technology of China, legal professionals, and the director of the Artificial Intelligence Governance Research Center of Tsinghua University. Different experts have cited different regulations, and they have different opinions on whether copyright and portrait rights are violated, which also proves that the development of AI technology is a new challenge for legislation and law enforcement agencies around the world.

Open Bilibili and search for "AI singer", we can still see various videos of "AI Stefanie Sun" covers and tutorials,

At least on the premise of "doing no evil", we who have not yet been subjected to strong "copyright guns" should take advantage of the moment and let "AI Stefanie Sun" sing another song by Jay Chou.