text | Linghu Boguang
Since the advent of chatgpt last year, generative AI has replaced this industry, and the industry has been in trouble for a while, but then it gradually subsided. At the beginning of this year, a number of new large AI models have emerged: sora, which generates text videos, kimi, which supports text processing of 2 million words long, and suno, the rising star of Vincent Audio.
What’s so scary about it? In the past, AI generated only fragments and not the whole song. Now with the suno music generation model, no matter whether the user has music theory knowledge or not, they only need to enter simple prompt words, such as music style, music genre, lyrics content, Tone, etc., wait a few seconds to quickly generate 2-minute music with lyrics and beats.
Recently, suno company also launched the v3 version, which can generate broadcast-quality music for the first time, and adds richer music styles and genre options, such as classical music, jazz, hiphop, electronic and other new trends. The official stated that the v4 version is already under development and plans to launch new features.
As soon as this news came out, it was natural that it was "chatgpt in the music industry", "subverting the music industry", "musicians will lose their jobs", etc. These were the comments made by countless netizens and media.
Combined with the already festering Chinese music scene, wouldn't it mean that Chinese music is the first to die? How terrible is
ai music generation? No, but it should not be underestimated.
I have not personally experienced this software, but I have watched some bloggers’ videos. After listening to the songs in the videos, I can basically judge that this AI algorithm is just a set of chords to make melody and off-key techniques. , the practice samples are basically European and American music libraries. If the practice samples are expanded to include music of various nationalities and styles, it can only be said that musicians who rely on chords to write songs will really not be able to make a living, and the future of music will probably only be able to accommodate first-class artists. How to say
? We can refer to the history of automobiles replacing horse-drawn carriages, cameras replacing paintings, film and television dramas replacing theater and other industries. In fact, it is difficult to compare horse-drawn carriages and cars, because this is a transportation industry that does not have much cultural value, and the latter is more suitable. If you look at cameras and movies, they have not completely replaced the old industries, but it is certain that the old industries have been greatly reduced.
At the same time, the gameplay becomes highly specialized or abstract. After all, new industries (movies and pop music have a short dissemination time), so the long-term gameplay and abstract art of the old industries cannot compete with traditional industries.
It did not completely replace it, but it did occupy people's lives. In the future, it will be more difficult to define plagiarism in popular music. With this set of tools, independent musicians can actually transform into soundtracks for cultural products. One person and one computer can complete the entire soundtrack for games, movies, and even radio dramas, comics, and novels. Although I earn less than before, I still have a way to survive.
I am an amateur music lover, and I also like it. When we say that the music scene is bad, we mean that it is all routines, that is, universal chords. It does not refer to a certain piece of music, nor can it be regarded as a universal chord for a type of music. This term describes a polyphonic writing technique. The important feature is that the same theme (which may also have changes) is repeated at intervals of the same beat. put.
The best example I can think of here is Mozart's "Pinking of Chrysanthemums", which is obviously a six-part canon and is very delicately written. In the process of backward development, canon rarely appears alone, and usually serves as the beginning of an insert or a development part.
I very much agree with someone who said that "pop music is relatively simple". In fact, this is quite consistent with the level of knowledge I perceive. On a deeper level, I think the fundamental reason is that the human ear is not designed to recognize things that are too complex. Nowadays, everyone thinks that AI has replaced music. The bias is that this model is only applied to pop music, and it is in the Chinese music scene. If you want to summarize "music" with data like
, you still have some clues. Because pop music is only a small part of music.For example, look at ancient times, people like Mahler and Scriabin, look at modern times, people like Stravinsky, Schnittke, and look at modern times, Penderecki, Toshio Hosokawa, etc. Man, attempts or research like this should be able to express music in more diverse ways. As soon as
ai came out, music became a routine? Pop music is indeed becoming more and more "routine".
The predecessor of pop music is more like a hybrid of minimalist music and jazz styles (of course, this may not be completely summarized). It has inherited some of the things explored by its predecessors. Things also introduce new concepts. For example, "traffic". This leaves it to the public to decide where pop music will go.
But is it really that something as profound as modernist music cannot be appreciated by the public?
I don’t think so. An example is the development of electronic music. When electronic music appeared in the 1930s and 1940s, it was the most avant-garde wave. By the 1960s, a machine as big as a pipe organ was still a novelty when it appeared next to Bernstein's conductor's podium. But we have become accustomed to it.
Using this technology to create new synthesizer sounds, or using this technology to simulate existing sounds, are commonly used nowadays. In addition to these, in the category of harmony, distant functional harmony has become commonplace in popular music, which is something that did not exist in the previous classical period or the earlier Baroque period.
So I think that the rich techniques in modern music will become part of the development of popular music. As a young person, I also want to share my ideas with others. There is also a critical element among these ideas. Whether correct or not, these ideas will be meaningful and valuable.
What we call routine chords, that is, the evaluation criteria for average songs are only part of the song evaluation criteria, that is, user expectations violate the model - the closer it is to the average song, it means that the song better meets the user's expectations, and the better the user can guess it It’s a double-edged sword as to what comes next in the song.
Users can guess the next melody, which means that the song falls into a cliche, but it is not a good thing if the song does not meet the audience's expectations.
Therefore, this AI model is often combined with information entropy, which is the expectation violation-information entropy model. In 2016, eerola proposed this model in the paper "expectancy-violation and information-theoretic models of melodic complexity".
At the same time, the idea of "average songs" itself is not new. In 1984, Simonton also conducted a content analysis of 15,618 classical themes in "melodic structure and note transition probabilities: a content analysis of 15,618 classical themes" , the melodic structure and note transition probabilities of these themes were studied. In
matlab, based on this content analysis, existing melodies and the average level of 15,618 classical compositions are analyzed to obtain a score, which can be used to evaluate the originality of the song, that is, in the 80s way of thinking , by comparing "average songs", it can only prove the originality of the song, but cannot deduce the "quality" or "goodness" of the song.
ai can’t replace pop music for the time being? If you don’t believe it, especially Chinese music
the reason is that the Chinese music scene is still very backward. Mean Song is a very interesting application. It uses fast Fourier transform to analyze the music waveform and directly process the waveform, so it is equivalent to all the parameters of the music. All integrated into it. In other words, using the black box of deep learning neural network to perform unified and undifferentiated identification, the resulting average song theory is an average in terms of rhythm, mode, listening effect, etc.
But music has different weights on emotions. In the black box of a neural network, you don’t know which one has a higher weight: rhythm, orchestration, bpm, or the singer’s voice.
For example, if Stefanie Sun had many songs in 2001, then maybe Stefanie Sun's voice would be the standard for average songs in front of the neural network, and the timbre of Stefanie Sun's own voice cannot affect the emotion of too many songs.So the fast Fourier transform of the waveform also has shortcomings. Better still, you should use midi directly for analysis, but it is difficult to build a large number of popular music midi libraries.
At the same time, the end of the golden age of the Jay Chou generation of Chinese music can be attributed to the simplistic logic of music that is slightly pale. The "chorus-ification" of the Chinese music scene depends more on the mentality of the audience and the streaming media they usually come into contact with. The birth of AI will affect music creation to a certain extent, but just like painting, it will not completely lead to the end of music creation.
Starting from the simplification of music, the article uses Canon chords as an example to describe how most songs use the same harmony routines and chord progressions, but do not take into account changes in orchestration, musical form structure, etc.
Finally, Chinese pop music is still very backward. We often have the feeling that the music is like Liu Cixin's "Poetry Cloud", where all the melodies and routines have been written out. In fact, European and American music styles alone include classical, jazz, rock, punk, r&b, house, britpop, trip-hop, gangsta, rap, synth pop, orchestra, chamber pop, folk, bossa nova, classical etc.
Chinese pop music focused on these aspects in the past few decades. In the first thirty years in the mainland, local folk songs, ballads, operas and red songs (and their popularity) were developed. Cui Jian pioneered Chinese rock, and Gao Xiaosong and others pioneered mainland folk songs; Hong Kong has Cantonese folk songs, opera, pop rock, and then all kinds of Cantonese pop.
The first thirty years of Taiwan’s music overlapped a lot with mainland music. In the 1970s, the folk song movement opened up Taiwan’s new folk songs, new folk songs, and Hokkien folk songs, followed by Luo Dayou, Qi and Qin’s pop rock, pop folk songs, and then pop, one-vote music Everyone plays this. Finally, the Jay Chou generation developed Chinese rb, rap, new Chinese style, and then fusion style.
may not be comprehensive. If you know a little about the world music scene, you will find that Chinese music styles are very rare.
This is like the popular Jay Chou and Golden Age, but among the so-called music lovers, do they often criticize "Hong Kong and Taiwan Guava" and "urban love songs"? Although it is absolutely true, it also proves that the music style of the Chinese music scene is very There are a lot of independent musicians and bands that are scarce, and people do play well, but not many people listen to them.
And if Chinese music learns from European and American music styles and does it well, it will be very different from European and American routine songs to a large extent. This is integration and localization, which is what Japanese music does.
You have to say that AI and its predecessors have done everything in melody, arrangement and style, and you can't do it without similarity and plagiarism.
Believe me, if Chinese-speaking audiences can hear it, it won’t mean that the Chinese-speaking music scene is dead. To put it simply, the same melody is changed to the rhythm, the same arrangement is changed to the orchestration, combined with the difference between Chinese and foreign singing methods, how many ordinary people can distinguish this? Not to mention, there are different expressions brought about by cultural context.
As long as plagiarism is not committed, this is how it works in any cultural industry!
