Author丨Xiao Xiao
Editor丨Wang Jun
The music industry has been "earthquakes" this year: first, in early March, Suno, known as the "music version of ChatGPT", was opened for free, and ordinary people could generate complete two-minute songs in a matter of seconds. Then, on April 2, more than 200 of the world's leading musicians issued an open letter opposing the misuse of AI, demanding that tech companies commit not to develop, destroy, or replace AI for human artists.
The letter was signed by 246 well-known artists in Europe and the United States, including Billie Eilish, who recently won the Oscar for Best Original Song, Katy Perry, Imagine Dragons, and the legacies of the deceased music giants.
In the open letter, the artists stressed that they are not completely negative about AI, "When used responsibly, we believe that this technology has great potential to enhance human creativity." But they believe that some platforms and developers are using AI to weaken musicians, and that predatory use of AI must be prevented, which could pose a huge threat to creators' ability to protect their privacy, identity, music and livelihoods, and in the long term slide the entire music industry.
Ding Tao, director of the Entertainment and Sports Law Research Center of Zhuojian Law Firm, told reporters, "The act of generating music often involves multiple rights subjects, such as songwriters, audio and video producers, and performers. In the development of AI technology, how to obtain authorization and how to distribute benefits are currently neglected. However, many interviewees also pointed out that at present, AI music has a greater impact on low-cost soundtracks, and it is still far from creating songs with artistic value and commercial value.
It is worth mentioning that in March this year, Tennessee, a strong music state in the United States, passed the "Ensuring the Safety of Sound and Image Similarity Act", which prohibits the use of AI to imitate the voice of artists without permission. The bill will go into effect on July 1.
AI music steals jobs?
Before the joint letter, AI music had been popular for more than half a year, and the most out-of-the-circle content was undoubtedly imitating the singer's voice. Many cover videos of "AI Yanzi Sun" and "AI Jay Chou" have exceeded one million hits on station B, and some foreign users have synthesized AI songs with the voices of Drake and The Weekend, which have been spread more than 10 million times on TikTok.
However, the self-entertainment of netizens is usually not enough to move the cake of musicians, and the impact of AI technology at the company level is much greater.
In particular, the letter singles out "some of the largest and most powerful companies that use artists' work to train AI models without permission, replacing the work of human artists with a large number of AI-created 'voices'".
Since the beginning of this year, various companies have rushed to launch AI music generation tools. After the release of Suno V3 on March 22, it quickly fermented and became a phenomenal application. Just by entering a simple text prompt, Suno can generate music with melody and lyrics for up to two minutes, and you can also fill in the lyrics by yourself, choose the music genre, and even some netizens use the manual of Lianhua Qingwen Capsule to generate a complete song with a prelude, chorus, and Rap.
The data shows that Suno had nearly 10 million visits in the first week of April. Music producer Wang Miao told the 21st Century Business Herald reporter that Suno's appearance really surprised the industry. Compared with similar products, Suno has made leaps and bounds, and the music it generates can already meet the music hearing and basic aesthetic needs of most people.
"In fact, the industry has always used some AI utilities, such as mixing processors powered by AI technology. In the past few years, generative AI has emerged, and everyone has more or less begun to use AI models to assist in creation and improve our creative efficiency. For example, Wang Miao said that in the process of music production this year, due to the problem of the singer-songwriter's schedule, there was a flawed vocal material that could not be re-recorded, so he used AI to clone the voice to replace the flaw. There are also peers who use AI voices to complete the singing of the entire finished song, and even some AI soundtracks have been put into commercial use in advertisements and TV programs.
The reason why you choose to use AI over professional musicians is usually because of your budget. According to Wang Miao's observation, this is mainly aimed at low-cost and low-expectation scenarios, such as for low-cost film and television, games and self-media creators, AI music can provide a low-cost BGM (background music) solution, which plays a role in reducing costs and increasing efficiency.
However, in order to achieve a certain artistic effect, it still requires a lot of manpower and a lot of cost. Last year, singer Chen Shanni collaborated on an AI project that attracted a lot of attention, and she used AI to sing a song called "Teach Me How to Be Your Lover", which successfully hid the mixing and mastering engineers, and even the audience. But behind this is the year-long training of the AI model, as well as Chen Shanni's own adjustment of the AI's singing style, articulation, tone, and mood of each word of the AI.
Wang Miao told reporters that there is still a huge gap between AI music and commercial distribution. At present, AI can only complete some repetitive and regular work, and the human-centered creation model has not changed, and it is difficult for professional musicians to be replaced.
The distribution of copyright interests is urgent
Compared with grabbing jobs, a more pressing issue for musicians and brokerage companies is copyright. The letter accuses the AI model of training it for significantly diluting the royalties paid to artists, preventing them from being paid fairly.
Ding Tao told the 21st Century Business Herald that there are two thorny infringement disputes behind AI music to be resolved: the first is the risk of infringement of training data that exists in both Wenshengwen and Wenshengtu large models. Training a large model requires a large number of musical products to be extracted and processed as materials, and generating music often involves multiple rights subjects, such as songwriters, audio and video producers, and performers, but there is not much discussion on how to obtain authorization and how to distribute copyright income.
In addition, it is difficult to avoid a certain similarity between AI music generators and training materials, but whether it has reached the level of infringement, Ding Tao said that it is necessary to discuss each case according to the principle of "contact and substantial similarity".
To a certain extent, Suno avoids the risk of infringement of music creations, and users cannot use the artist's name as a prompt. For example, if a reporter wants to generate a Jay Chou-style song, and after typing in "Jay Chou" or "Jay Chou style", Suno will classify it as R&B. The product is closer to a general R&B song, and you can't hear the shadow of a certain Jay Chou song, let alone mass produce Jay Chou-style songs.
Image: Suno Prompt "Describe the style of music you want, the Suno model does not recognize the artist's name, but understands the genre and style. ”
However, some netizens pointed out that if the style of the prompt words is niche enough, coupled with the continuous and precise adjustment of the prompt words, Suno can still generate a music with a strong "sense of splicing" - which means that it is easy to hear which tunes the AI specifically disassembles and respells.
Ding Tao said that it is difficult for both situations to constitute infringement. This is because copyright does not protect "ideas" such as style and theme, but only specific expressions. Even if you can hear that individual parts are the same as the original song, it is difficult to meet the standard of "substantial similarity" in copyright without affecting the whole song.
"Out of respect for original industry practice, even a song melody that only takes a few seconds should pay for the copyright license. However, this can only be regarded as a self-discipline norm and cannot be used as a basis for judging whether there is infringement. Ding Tao said.
Industry friction intensifies
It is worth noting that the contradiction between musicians, copyright owners, and Internet platforms to share the cake has been going on for a long time, and the AI problem has added a new sum to the layers of old accounts.
In January, Universal Music directly condemned TikTok in an open letter for "making Universal Music accept a deal that is far below fair market value." The spearhead is that Tiktok refuses to share revenue and only accepts a one-time copyright transaction.
This led to a rupture in the relationship between Universal Music and TikTok, with Universal Music removing songs from TikTok that included globally renowned artists such as Taylor Swift, and a large number of short videos using the soundtracks of these songs were "muted." (See: Global Douyin "Mute" Universal Music, Copyright Game Escalation)
There are also a number of musicians in this objection letter who have partnerships with Universal Music. As one of the world's top three record labels and a company that manages nearly 4 million songs in its rights agency business, Universal Music has repeatedly said that it has been troubled by AI in a year. For example, in an open letter against Tiktok, Universal Music accused TikTok of allowing the platform to flood the platform with AI-generated songs, and supporting, promoting, and encouraging AI music creation, diluting the copyright pool of creators.
Not only Tiktok, but also various social platforms and music streamers are developing their own AI tools: Spotify has recently released AI DJ, AI automatically generated playlists and other functions, and domestic NetEase Cloud Music has launched one-stop AI music creation tools such as NetEase Tianyin, and cooperated with CCTV news special programs to AI soundtracks...... Although artists have jointly called on technology companies not to develop, destroy or replace AI for human artists, it is clear that it will be difficult to stop the pace of technology, and there will only be more friction in the future.
In addition, the open letter also mentioned the impact of AI music on destroying the industry ecology and lowering industry standards, and there are optimistic and pessimistic attitudes towards this.
Ding Tao is worried that although AI cannot replace human creative arts, it can produce low-quality "saliva songs" with very low cost and mass production, and ferment them through social media, which may lead to the end of bad money driving out good money.
Wang Miao has a different view. "AI music models like Suno are now more of an entertainment product. For more people, it's just a fun look at the excitement. From a longer-term perspective, Wang believes that AI creation of regular and common works is indeed more efficient than humans, and this process will inevitably reduce production costs and have an impact on musicians with limited aesthetic and technical skills. However, at the same time, the more unique it is, the less likely it is to be replaced by AI, so it will gradually optimize the division of labor in the industry.
"Will it be the starting point of the decline of the music industry, if the rules and benefits are well distributed, musicians can coexist well with AI. The people who can stand firm in the industry in the future must be very good people who can control AI well. Wang Miao said.
SFC
Editor: Li Yutong, Intern: Huang Lihong
The China Securities Regulatory Commission has blockbuster new regulations
The State Council released a blockbuster, nine highlights! Wu Qing spoke
The price suddenly skyrocketed by 10 times!