text | Wu Daodun, author | Summer
Time games are always filled with many wonderful "coincidences".
In 1787, Mozart invented a musical dice game. The game props are 176 minuet bars, 96 trio bars, two rule sheets filled with numbers, and two dice. The rules of the game are to randomly throw two dice 16 times. According to the throwing numbers, select the corresponding music fragments in the rule table to form a random minuet.
leaves the final arrangement of the music to some kind of rules. Mozart may have done this just out of curiosity and entertainment, but the "machine selection" that this entertainment and curiosity inadvertently touched will lead music to a completely different direction 237 years later. New Kingdom.
Now, suno has the key to open the door.
ai transforms music, and
ai transforms music from 2.0 to 3.0. It seems to be a trend that has only become popular in the past two years. But its technology has been settling in for some time.
In 1957, two professors, lejaren hiller and leonard issacson, believed that the beauty of musical works is because they conform to certain rules, and if a computer can learn such rules, it can also create beautiful music. With this idea in mind,
used program settings such as "the range cannot exceed one octave and must start and end on the C note" to create the string quartet "Iliac Suite", completing the first piece of music in human history. A piece of music "composed" entirely by a computer.
Whether you agree that this set of computer-created music is artistic, it can at least be regarded as a complete exercise. "Is music an expression of emotion or rules? Can computer-generated music be called music?..." Discussions surrounding artificial intelligence and music have continued. People have foreseen the possible threats that artificial intelligence may bring, but strongly deny it. existence of a threat.
After this, ai music developed rapidly.
In 1986, David Cope, the godfather of AI composition, created the music software emi. In 1997, a professor played three piano pieces to the audience at a lecture at Stanford University, one by Bach, one by Emi imitating Bach's style, and one by the professor's husband, asking the audience to guess. Whose are these three pieces of music? Most people regard the professor's husband's works as computer works, and Emi's works as Bach's works.
Of course, this may be because the professor’s husband’s composition skills are not proficient enough, or the average listener’s sensitivity to music is not high.
In 2016, a French computer scientist decided to create a musical artificial intelligence model after watching the science fiction movie "Her", and aiva was born. Through in-depth learning of 15,000 pieces of music by famous artists such as Mozart, Bach, and Beethoven, aiva began to create its own musical works.
When the team mixed aiva's works with those of the composer, the invited musicians could barely tell the difference. Later, aiva became the first non-human member of the "French and Luxembourg Composers Society" (sacem) and had the right to authorize his own works. Before
, people were still debating whether music created by machines could be called music. Now, people are trying to make peace with it and thinking about making money from it.
aiva mainly targets industries with a large demand for classical music such as games, advertising, and movies. It has participated in the creation of background music for important events such as the opening ceremony of Luxembourg National Day and the NVIDIA GTC 2020 Conference.
At the same time, aiva also provides customized song services to individuals, such as assisting pop singer taryn southern in completing the creation of "Love Sick".
is similar to aiva. The subsequent commercialization of the AI music model will mainly focus on the two directions of video soundtrack and auxiliary singer creation.
At this time, AI has been able to create complete music, but the music created entirely by it is too modular. To make the music artistic, humans still need to screen and tune it. AI exists more as a tool to assist humans in artistic creation.
People marvel at AI's regularity in music processing, and are thankful that it is slow in "understanding" the artistry of music.
chatgpt moments in the music industry
When suno appeared, everything was different again.
Recently, the AI music generation software suno released the v3 version. With only a simple description, it can generate broadcast-quality music that lasts about 2 minutes and includes lyrics, intro, interlude, and chorus.
In addition, suno also supports four musical instruments: drum, piano, violin, and cello; 29 music styles such as relaxation, psychedelic, and atmospheric; music generation of 71 music genres such as folk, funk, and heavy metal, etc., and the scope is basically covered mainstream music aesthetics. The strength of
suno is audio production. In terms of text, chatgpt can be called to generate lyrics and titles. On the AI music ranking list on the suno official website, many songs have been played over 10,000 times. We also tried suno for music creation. It’s hard to say whether
will win the award, but it’s more than enough to replace the popular Internet divinity. If we agree that saliva songs are artistic in a broad sense, then we have to admit that AI can "reproduce" art.
Similar to suno, NetEase Cloud Music’s AI music creation platform NetEase Tianyin will be launched in April 2022. NetEase Tianyin is divided into two modes: AI song writing and AI arrangement. The normal mode of AI song writing, which is similar to suno, is aimed at pure beginners with no knowledge of music theory. The custom mode of AI arrangement, which is similar to suno, is aimed at people with a certain level of music. Basic creator. In
novice mode, the operation methods of both are very simple. NetEase Tianyin can even provide more personalized services such as vocals and accompaniment selection. However, judging from the generation results, the music artifacts generated by suno are weaker.
Last March, people had already foreseen the potential of artificial intelligence to transform the music industry through chatgpt. But at that time, the research on AI by all parties was still limited to general large-scale models. Everyone in the music industry optimistically estimated that it would take several years for AI to transform the music industry. Then just one year passed and the countdown began.
Obviously, what is more worrying than AI music itself is the development speed of AI music.
's current suno can generate relatively high-quality saliva songs after a few hours of training. Put these songs together with the saliva songs that Shenqu Production Company spent several days to produce on the short video platform. It is difficult to say which one of them will be more popular. The monthly package of the commercial version of
suno is only US$24 (annual payment), which is equivalent to less than 200 yuan, and the monthly salary expenditure of an employee starts at thousands. Music companies still need people to train the music produced by AI, but There is no need for so many people to do the same job.
Those who were kicked out either changed careers or engaged in music creation that required higher artistic requirements.
Like chatgpt, suno cannot escape data constraints, but the data pressure it faces is much greater than chatgpt. The metadata of
music is more complex than images and text. The midi data containing most of the creation details retained during the music production process belongs to the copyright owner. Technology companies cannot obtain it from the Internet. Without original data, the training effect will be greatly reduced.
The three major record labels, which have data and funds, are cursing short video platforms and music platforms for encouraging the development of AI music and infringing on the rights of musicians, while at the same time stepping up research and investment in AI music. established sonyai with Sony's Sony Music, Warner acquired asodatone and invested in lifescore, and Universal invested in soundful.
When the founder of aiva was asked why he chose to focus on classical music, he said that the classical music database used to train aiva did not involve copyright issues and the copyrights had expired. Now it seems that he was quite prescient.
Musicians, can they find a way out from art to music?
The control of record data by the three major records will delay the development of AI music models by platforms and technology companies, buying more time for musicians; however, the lack of data may also force platforms and technology companies to continuously optimize their own models. It makes the moment of singularity arrive earlier and accelerates the unemployment of musicians.
No matter what, this huge unemployment crisis cannot be avoided. Fortunately, the main battlefield of AI is in video soundtracks, advertising music and other functions, and in art and music, musicians can still get a chance to breathe.
Ordinary advertising soundtracks and short video soundtracks can be completed by AI, while professional movie and game soundtracks may be more appropriately left to professionals. First, such customers have higher artistic requirements for music, and qualified soundtracks cannot To satisfy their needs, they need to create surprises.
Secondly, it is more convenient for customers to directly hand over the modifications to humans for detailed requirements such as the type of orchestration and note duration in a certain measure of music. In the
movie "Simone", a virtual program has become a highly sought-after new film and television star after careful marketing and packaging. Although this may be due to the agent hiding Simone's virtual program. But in the era of AI music, even if people know Simone's true identity, they may not not love her.
Similarly, who knows whether the music industry will give birth to its own Simone. After all, virtual singer Hatsune Miku’s concert appeal is as good as that of a real person. As virtual reality technology continues to develop, it is only a matter of time before the price of VR glasses drops. Cyber music festivals may become a new fashion.
When many listeners are impressed by the music created by AI, they often add the sentence "AI music always lacks a little emotion." Although people listen to music just for the music itself, the stories behind the music can give listeners more room for emotional reverie. As long as the eggs are delicious, why care about the chicken that laid them?
In the era of AI music, eggs and the chickens that lay them will become equally important. In addition to musical works, increase the personal exposure and marketing of musicians, and add stories other than music (these stories usually exist in themselves, but have not been discovered).
Don’t feel morally guilty and think that you are using music to “bring in private goods.” In an era when the public is still uncertain about whether AI is a creation or an imitation of art, music created for human beings through “bringing in private goods” There's nothing heinous about recruiting a larger audience.
Of course, you can also continue to be a music fundamentalist and only talk about music and nothing else. In an era when everyone is moved by AI, your persistence and existence are a kind of strength in itself.
Compared with these, we may be more worried about human beings losing their dominance in the cultural field.
Now, AI has been able to simulate low-level aesthetic emotional expressions in musical works. In time, AI may be able to simulate deeper and more complex aesthetic emotional expressions in musical works, creating Mozart No. 1 and Mozart that are comparable to Mozart’s musical works. No. 2...
Although from the perspective of "objectification of human essential power", these Mozart No. 1 and Mozart No. 2 can never be called works of art, but they are enough to satisfy human aesthetic needs together with Mozart's works. .
Today, musicians are moved by Nirvana, Dou Wei, and Wanqing, and create music along their spiritual direction.
Tomorrow, musicians will be moved by Nirvana No. 1, Dou Wei No. 1, and Wanqing No. 1 created by AI, and they will create based on the human spiritual beliefs that AI musicians understand.
n years later, Mozart No. 1,000,000 disappeared and was replaced by Artificial Intelligence No. 1.