Is AI Stefanie Sun the hottest Chinese singer in 2023?

Some fans say that when encountering a song they like, they can be performed by their idol's voice, and they can also watch and listen to it for free, which is a "good experience".

2919 words in full text, reading takes approximately 6 minutes

New Beijing News reporter Zhang Jianlin, editor Bai Shuang, proofreading Li Lijun

Players familiar with AI music production have heard of AIVA to some extent. This virtual composer, founded by AIVA Technologies in Luxembourg and the first to receive international certification, has been in high demand since its birth in 2016. Some music producers have discovered the strengths of this virtual composer: AIVA has learned a lot of music works, built a network model based on the musical characteristics of the songs, and ultimately relied on the model to assist musicians in producing the songs they want.

AI music generators are also starting to appear on more ordinary users' computer desktops. After undergoing a series of operations such as AI voice changing and text to song conversion, the new music works are exported with one click. So, fans who failed to promote idol's "business" have a new hope - the appearance of "AI singer". Fans have covered many classic songs according to their own needs. Nowadays, "AI singers" are producing songs faster and have good traffic.

The cover work of "AI Stefanie Sun" uploaded by netizens. Website screenshot

Producer: The cleaner the sound material, the better the inferred sound quality

UP host Yingtai Li (pseudonym) posted 36 videos on Bilibili. The most popular one among them is "AI Stefanie Sun" covering a song by Jay Chou, with nearly 1.25 million views played. A netizen commented on the video, "The hottest Chinese singer in 2023 is AI Stefanie Sun

Before updating AI Stefanie Sun, her most popular video was a game character facial capture effect created using LiveLinkFace (a real-time facial animation software for virtual production), with 136000 views.

Currently, the videos released by Intelli mainly feature songs covered by "AI Stefanie Sun". She will add a disclaimer below each posted video. It reads: This work is only released for entertainment purposes, and the possible consequences are not related to the author or contributor of the audio conversion project used. This video uses AI synthesis technology, and the singer's voice in the video is not Sun Yanzi herself. If there is any infringement, please inform us to delete it.

Creating AI music is about listening to the feeling that one likes singers singing other classic songs. The collision between the two can have different effects, "Yingtai Li said in an interview with the Beijing News." But now, when AI changes the sound, the only way to do it is to change the tone, and the singing of the song really doesn't have much emotion

She told reporters that each song produced using AI takes a different amount of time, depending on the producer's needs and level of dedication. AI composition has lowered the threshold for the general public to create music, but in reality, it is not easy to create a complete AI piece.

Mo Xiao (pseudonym), a UP host who teaches netizens to "train AI singers" on Bilibili, said that the core technology comes from open-source projects such as So-VITS-SVC and RVC.

According to him, when downloading an open-source sound AI software material package, the producer first needs to process the singer's relevant sound data, and then use the obtained audio data to create a training acoustic model. Generally, the longer the training time, the better the effect. The model is used for reasoning and replacement, and after some text is converted to speech, audio editing software is also used for post synthesis.

AI voice change is an important part of the entire process of producing cover music for "AI singers". Mo Xiao mentioned that the material of sound data is very important. The cleaner the singer's voice, pronunciation, breath, and other materials, the better the inferred timbre will be.

In March of this year, singer songwriter Chen Shanni released her new song "Teach me how to be your lover". In an interview with a reporter from the Beijing News earlier, she stated that in this song, every sound, every breath, including all harmonies, are all performed in AI. The 'singer' is an AI model trained by herself, and the single cover is also generated by AI. In her opinion, whether AI has emotions or not depends on whether the listener thinks it (the song produced by AI) has emotions or not.

Mo Xiao believes that "as an auxiliary technical means, the main problem with using AI sound transformers and other software to train sound is that it cannot feel and understand the emotions of ordinary people. This method only discovers patterns through extensive training of songs, imitates them, and forms rhythm, which is definitely different from the actual singing of the person

Fans: The song they like is performed by their idol's voice, and the experience is good

Intelli believes that listening to AI music may be out of curiosity about new things, or it may be because the singer's "fans" want to hear the singer sing different styles of songs.

A week ago, Mr. Chen from Wuhan learned about "AI Sun Yanzi". On the way to work and squeeze the subway, he used his phone to swipe short videos and accidentally caught a video of Stefanie Sun singing someone else's song. He didn't look carefully, but as a fan of Stefanie Sun, after listening to a few words, he felt a bit strange. Later, after watching the introduction in the video, he realized that this was not done by Stefanie Sun herself, but by AI.

There are differences, but overall it feels okay, "Mr. Chen said

Mr. Chen doesn't mind if these covers are sung by AI rather than Stefanie Sun herself. When encountering a song he likes, being able to be performed by his idol's voice, and being able to watch and listen to it for free, he feels that the experience is good.

He told reporters that he had little thought about whether the people who made these music would infringe on relevant copyright, portrait rights, etc. He believes that these short video platforms generally have a content review system, which can be publicly available and can "survive" for so long, so there is probably no problem. Unless they (AI music producers) use the produced works for commercial purposes.

However, Mr. Chen also has his own concerns. AI is really powerful, and I'm worried that my idol won't sing, and there are AI versions of her singing other people's songs online. So, I'm wondering if this will give songwriters some warning that without good works, they may be replaced by AI versions of themselves

In addition to the popular "AI Stefanie Sun", on Station B, Cyndi Wang, Faye Wong, LanaDelRey (American pop singer) and other popular artists have a large number of AI cover songs based on them. Many listeners who listen to the cover of "AI singers" will willingly accept their favorite artists singing others' songs in AI format.

A "fan" of Yao Beina left a message under the video of the "AI Stefanie Sun" cover work, "I am a shell fan (referred to as Yao Beina fan). She went to heaven, and I really hope someone can create an AI Yao Beina and let fans listen to her cover songs again

Music enthusiasts: AI music cannot surpass top art creations

Mo Xiao told reporters that they will create new songs based on comments and comments. The vast majority of netizens want to console themselves with songs covered by their idols. Even if they are not the idols themselves, their voice tone is similar, which can still satisfy fans.

Beijing music enthusiast Xiao Zhou is also a listener of "AI Singer". He believes that AI has surprised the public in creating fresh music products. In the early days, AI music creation was limited to composition, but now singers' voices can be simulated and mixed to create music, which feels very interesting.

The reason why we say 'AI singers' are not like ourselves is because artists created by artificial intelligence lack human emotional experience and are not real emotional expressions. Secondly, we do not fully understand the working mechanism of the human brain in artistic creation, and AI technology cannot currently do it. In his view,' AI singers' and music produced by AI cannot truly surpass top artists and their works.

Zhang, a student majoring in recording art at Wuhan Conservatory of Music, believes that imitating music through AI sound production can basically achieve the effect of sounding "very similar" to an artist's voice. But at the same time, the sound synthesized by technology is far less attractive than the sound rich in on-site emotions.

A listener of AI music believes that art is not technology, and in the field of art, AI can only serve as an auxiliary, but not the protagonist. In her opinion, although artificial intelligence has strong learning and imitation abilities, it cannot learn the essence of music as a product of human spirit.

Why do we go to concerts? Singing is an act. A living person can bring empathy and infection through the act of singing, and even different interpretations such as expressions, atmosphere, and improvisation. She believes that art and social life are matched, and experience is the first step to art. Therefore, music is not simply music, it is important to express people's inner things.

On duty editor Li Jiajian Kang Xixi

Recently revised WeChat

Readers and friends often miss push notifications

Star logo "New Beijing News"

Timely receive the latest and hottest tweets

Click on 'watching' to share hotspots