Music artificial intelligence draws attention to creation to meet general appreciation needs

  Music creation lowers the threshold, AI composes sad feelings

  In just a few short years, artificial intelligence in music has become a hit.

Composition, accompaniment, improvisation... The "talent" displayed by artificial intelligence on various occasions is eye-opening.

In the past 2021, in various academic activities such as the World Music Artificial Intelligence Conference, the Global Artificial Intelligence Technology Conference, and the 2021 China Music Technology Application Forum, music artificial intelligence has also become one of the most discussed and discussed topics. .

  What will the future of concerts look like?

Will musical artificial intelligence replace human artistic creation?

Will it give wings to human creativity, or will it be the terminator?


  AI girl learns music

  "Xiaoice has excellent learning ability"

  In the graduation season of 2020, the Shanghai Conservatory of Music welcomed a "special" graduate, the artificial intelligence girl Xiaobing.

Xiaobing not only graduated successfully, but was also awarded the title of "Honorary Graduate" of the 2020 Music Engineering Department.

  Talking about entering music content creation, Li Di, the former vice president of Microsoft (Asia) Internet Engineering Institute and the current CEO of Xiaoice, told the Beijing Youth Daily reporter that, unlike the previous poetry creation, visual creation, and singing, music creation has an impact on artificial intelligence. For Xiaobing, it is more difficult, "Music is streaming media, and the fluency of the work is very high. Previously, Xiaoice's ability in music creation was mainly content generation, such as singing. This time Xiaoice learned to write lyrics and compose music. And arranging is actually more difficult.”

  How to teach artificial intelligence to learn music creation?

The Xiaobing team and the teachers of the Shanghai Conservatory of Music told the Beijing Youth Daily reporter that the learning process for AI Xiaobing to start music creation is divided into three steps:

  First of all, based on Xiaobing's existing music creation model, the team gave the port to the teachers of Shanghai Conservatory of Music, and let the teachers listen to the massive music works created by Xiaobing and give systematic comments.

  Secondly, the teacher also taught the team programmers basic music theory knowledge, including arrangement skills, etc., to optimize Xiaoice's creative model.

  In the end, Xiaoice kept learning, handing in homework, and the teacher kept giving feedback.

No matter "model", "job" or "feedback", it is a continuous iterative process, and so on...

  When Xiaoice reaches a level similar to her human classmates and tends to stabilize, she graduates.

  During his studies at the Shanghai Conservatory of Music, Xiaobing accepted the "guidance" from Professor Yu Yang and Chen Shizhe, the dean of the Department of Music Engineering.

Talking about the teaching mode, Mr. Chen Shizhe said that teaching students to learn music creation is similar to teaching artificial intelligence: "In traditional teaching methods, we let students learn classic works, then copy, then teachers make corrections, and students give feedback. There is not much difference between the learning methods of intelligent and ordinary students. We also prepared some data for Xiaoice. After Xiaoice learns, he will create, and the teacher will evaluate it again to point out any problems. Continue to revise and complete it together. "

  Of course there are differences.

Chen Shizhe commented that "XiaoIce is a student with strong learning ability", "The biggest difference in the teaching process is that XiaoIce's iteration speed is completely unattainable for humans. For example, a student learns, copies a piece of music, and writes Homework, teacher revision, this process takes at least a week, but in the teaching of Xiaobing, we put hundreds or even more songs in one go."

  Teacher Chen Shizhe introduced that there are generally three horizontal lines for judging whether students can successfully pass the graduation assessment. Good ordinary pop songs; the basic line is to create commercials, TV series, documentary soundtracks, etc. After half a year of study, we believe that XiaoIce has reached the basic line of ordinary, simple song creation ability, and is constantly approaching the creation standard of the middle line, so we give She won the title of honorary graduate. In fact, her standard is very close to the middle line, and we also expect Xiaoice to continue to be admitted according to the standard of graduate students.”

  Li Di said, "For the industrialization of music, artificial intelligence can produce and supply content with stable quality, extremely low cost, and high concurrency without human participation. There is a lot of room for landing.”


  music artificial intelligence

  Create a complete song in 3 seconds

  The algorithm of artificial intelligence has entered the artistic creation, which has had a huge impact on the creation of traditional music.

In 2019, Huawei used the AI ​​in Mate 20 Pro to compose the remaining scores of Austrian composer Schubert's unfinished "Symphony No. 8" and performed it at a concert in London.

Unlike Schubert's unfinished draft, Beethoven's Tenth Symphony has almost no music score. Due to the deterioration of his condition, Beethoven only drew a few sketches of this work and died suddenly.

A group of scientists and musicologists from the creative artificial intelligence startup Playform have completed Beethoven's unfinished business. After more than two years of hard work, they not only let the artificial intelligence learn all Beethoven's works, but also taught it Beethoven's creative process. .

In October 2021, this artificial intelligence helped Beethoven continue to write the Tenth Symphony, which will be premiered in Bonn, Germany, attracting the attention of the world.

  "Like human learning, artificial intelligence also learns massive content through corpora, and then 'creates' new works." Facing the super learning ability of artificial intelligence, the director and professor of the Department of Music Artificial Intelligence and Music Information Technology of the Central Conservatory of Music Li Xiaobing couldn't help but sigh, "It's too strong!" In 2021, the "Theoretical Research on the Collaborative Innovation and Development of Music and Artificial Intelligence" declared by Li Xiaobing as the chief expert was selected as the 2021 National Social Science Fund Art Major Project.

This topic starts from the fields of musicology, brain science, artificial intelligence, computer science and other fields to carry out interdisciplinary research. "Intelligent Collaborative Innovation Development" and "Music Artificial Intelligence Philosophy - Aesthetics" are four sub-topics, focusing on the relationship between people, music and artificial intelligence, and conduct in-depth and comprehensive research on the problems in the collaborative development of music and artificial intelligence.

  With the deepening of research, music artificial intelligence is constantly improving in efficiency.

Li Xiaobing revealed that artificial intelligence can create a song in as little as 3 seconds.

"The original fastest time was 23 seconds, but now it has been shortened to 3 seconds. 3 seconds is not only for composition, but also for singing, accompaniment, and synthesis, which means that it is a complete work."

  Li Xiaobing and his team put together songs composed by humans and children's songs created by artificial intelligence, and invited volunteers to do a blind test. It was difficult to tell which song was composed by artificial intelligence.

Different from human thinking, Li Xiaobing used a "blind box" to describe artificial intelligence. "Some of the things created by artificial intelligence in music are the same as those of humans, but some are not the same, and even make people jaw-dropping."


  "The charm of music lies in the second creation of the performer"

  "The most difficult part of the combination of artificial intelligence and art is actually music." Yu Yang, a composer and director of the Music Engineering Department of the Shanghai Conservatory of Music, told the Beiqing Daily reporter, "It is easy to achieve if there are rules to follow, but music is time-consuming. Art, the auditory experience of people is infinitely variable, and there are too many uncertainties. Limited by technical level, data and other factors, the current music artificial intelligence is still in its infancy, and further exploration and research are needed.”

  Li Xiaobing also believes that compared with text and language, music has many more dimensions, including pitch, rhythm, strength, emotion, etc., and it is also much more complicated. Therefore, music artificial intelligence requires larger models and computing power, and also To the top scientists and artists hand in hand.

Although artificial intelligence composition can pass the Turing test by learning from a large number of works, imitate the creation of composers, and be accurate in performance, but people in the industry generally believe that in terms of emotion, these works are similar to those created by humans. There is a big difference.

  "That's the charm of music. A piano piece, when ten performers play it, is ten completely different feelings, and there is a second creation by the performers. This kind of difference in feeling cannot be replaced by a machine." Yu Yang said.

  Although it cannot replace the core creativity of human beings, artificial intelligence can already assist human beings to do many things, especially a lot of repetitive and basic work.

  "Twenty years ago, we all wrote things on the staves. Now most composers use computer software to create things, which is actually a form of artificial intelligence." Yu Yang, a composer, said that the writing was completed. Afterwards, the software can automatically generate performances, and composers can modify and adjust them at any time. "Artificial intelligence has greatly improved the efficiency of creation."

  Li Xiaobing also believes that "if you want to create high-level music, such as movie soundtracks or first-line pop songs that can hit the charts, artificial intelligence still has a long way to go, but for some background music, the creation of artificial intelligence has been able to It meets people's appreciation needs."

  With the in-depth development of music artificial intelligence, future concerts will undergo earth-shaking changes.

In Li Xiaobing's view, music in the future may not only be 3D music, robot hosting, interactive multimedia, artificial intelligence accompaniment, virtual reality, robot conductor, robot performance, robot singing, etc., but also new forms may appear. "Just like the emergence of cameras, The art world has been forced to undergo major changes, and a number of great artists that cameras cannot replace have emerged. With the deepening of music artificial intelligence research, future music creation will be more precious, and unexpected new music will surely be born. form."

  This edition of the article / reporter Tian Wanting Zhang Zhiyi

  Coordination/Photo courtesy of Liu Jianghua/Visual China

  Trend Archives

  Through its powerful digital processing capabilities, AI takes into account the whole process of songs from production to singing, realizes the simplicity of complexity, and completes the creation of musical works in a short period of time.

By learning how to write songs through algorithm learning and “practical” training, non-musicians can also use this technology to create their own unique songs.

  Industry layout

  Tech giants have their own priorities for investment

  Capital has long realized the commercial value of music artificial intelligence, and artificial intelligence technology giants are actively deploying: as early as 2016, ByteDance established AI Lab to conduct in-depth research on AI; NetEase Cloud Music strategically invested in 2020. AIVA, an AI music company, focuses on the field of AI-assisted music creation; Tencent AI Lab has launched the AI ​​digital human "Ai Ling", which can automatically generate lyrics and sing through keywords provided by users; the Tacotron-based acoustic model developed by ByteDance And the Chinese singing voice synthesis (SVS) system of the WaveRNN neural network vocoder - ByteSing, the AI ​​music open source project Muzic released by Microsoft Research, etc. are all trying to prove that artificial intelligence can complete the entire music production process.

  Discipline Reserve

  The first batch of music artificial intelligence Ph.D. graduates this year

  Major universities are also rapidly advancing the research and application of music artificial intelligence.

Central Conservatory of Music, Shanghai Conservatory of Music, Sichuan Conservatory of Music and other colleges and universities have opened music artificial intelligence majors.

In July 2022, the first batch of Ph.D.s in music artificial intelligence from the Central Conservatory of Music will graduate, and they have become talents that major Internet companies are competing for.

"Because the society has a huge demand for talents in music artificial intelligence, most of the current large Internet companies use them from the acoustics major, and they may not be specialized in music artificial intelligence. Students are very popular." Li Xiaobing said.

  "Usually, people who engage in music don't understand technology, and those who engage in technology don't understand music. Music artificial intelligence is an interdisciplinary field. What we are doing now is to combine these two aspects." Yu Yang revealed that the Shanghai Conservatory of Music's The Ph.D. in artificial intelligence adopts a dual-tutor system, with one tutor in music and one in technology.

"The future of artificial intelligence in music is promising, and it is very important to take this step now. But there is still a huge talent gap."



  The Shenzhen Symphony Orchestra performed the world's first AI symphony variation "My Motherland and Me", which is also the world premiere of the piece.

  July 9, 2020

  The theme song "Zhilian Home" sung by four artificial intelligence robots including Microsoft Xiaobing, Xiaomi Xiaoai, Baidu Xiaodu and Bilibili Lingyuan appeared at the opening ceremony of the 2020 World Artificial Intelligence Conference.

  October 9, 2021

  Beethoven's "Tenth Symphony" written by artificial intelligence was released in Germany and attracted worldwide attention.