Australian media article: Artificial intelligence continues to write Beethoven's Tenth Symphony

On September 24, the Australian "Dialogue" website published an article titled "How Musicologists and Computer Scientists Completed Beethoven's Unfinished Tenth Symphony" by Professor Ahmed Jamal, director of the Art and Artificial Intelligence Laboratory at Ruttgers University. Attempts were made to machine up Beethoven's creative process —how he wrote musical bars and painstakingly developed them into passionate symphonies, quartets, and sonatas. The full text is excerpted below:

When Ludwig von Beethoven died in 1827, he had only been writing the Ninth Symphony for three years, and many praised it as his masterpiece. He had already begun composing the Tenth Symphony, but had not made much progress due to deteriorating health. All that remained behind him were some musical drafts.

Too few drafts are left

Since then, Beethoven's fans and music theorists have painstakingly explored what the Tenth Symphony might look like and lamented the unfinished work. Delving into Beethoven's notes can yield some sort of wonderful reward, even though at one point it seemed like it would never be reachable.

Now, thanks to the efforts of a group of music historians, musicologists, composers and computer scientists, Beethoven's vision will become a reality.

I chaired the AI part of the project, leading a group of scientists from art platform AI companies. Scientists taught All of Beethoven's work and his creative process to a machine.

The complete recording of Beethoven's Tenth Symphony was released on October 9, 2021, the same day as the world premiere in Bonn, Germany. This is the result of more than two years of hard work.

Around 1817, the Royal Philharmonic Society of London commissioned Beethoven to compose his ninth and tenth symphonies. Symphonies composed for orchestras usually consist of four movements: the first movement is played at a fast pace, the second movement is slower, the third movement is medium or fast-paced, and the final movement is fast-paced.

Beethoven completed his Ninth Symphony in 1824. The symphony ends with the timeless Ode to Joy.

But when it comes to the Tenth Symphony, Beethoven didn't leave much behind, just some musical notes and hastily written down ideas.

In the past, there have been attempts to partially reconstruct Beethoven's Tenth Symphony. The most famous event was, in 1988, when musicologist Barry Cooper made a bold attempt to complete the first and second movements. He orchestrated together 250 musical bars from Beethoven's drafts to create the first movement that he believed to be faithful to Beethoven's vision.

However, Beethoven left too few drafts for symphony experts to reconstruct more.

Try something like never before

In early 2019, I was contacted by Matthias Leder, director of the Karajan Institute. The Karajan Institute is an institution that promotes music technology in Salzburg, Austria. Ledell explained that he was assembling a team to complete Beethoven's Tenth Symphony to celebrate the composer's 250th birthday. He knew I was working on artificial intelligence to create works of art, and he wondered if artificial intelligence could help fill the void left by Beethoven.

The challenges may seem daunting. To get this done, AI needs to do something it has never done before. But I said I would try.

Leder then assembled a team that included the Austrian composer Welter Verzova. Verzova, best known for composing Intel's signature advertising sound, is tasked with orchestrating a new type of work that will fuse what Beethoven left behind with what is generated by artificial intelligence. Computer music specialist Mark Gotham was responsible for transcribed Beethoven's musical drafts and worked on all of his works. In this way, the ARTIFICIAL can receive appropriate training.

The team also includes Harvard musicologist Robert Levine. Levin, who also happened to be a great pianist, had previously completed many incomplete works of the 18th century by Mozart and Johann Sebastian Bach.

In June 2019, the team held a two-day workshop at the Harvard University Music Library. In a large room were pianos, chalkboards, and a pile of Beethoven's notebooks—notebooks that included most of Beethoven's known works. In this room, we discuss how to turn fragments into complete musical compositions and how artificial intelligence can help solve this puzzle while remaining faithful to Beethoven's creative process and vision.

The music experts in the room were eager to learn more about the types of music that ai-powered had created in the past. I showed them how AI successfully generated Bach-style music. However, that was just a treatment for an input melody that sounded like Bach. This is a far cry from what we need to do. What we're going to do is reconstruct the whole symphony based on some phrases.

At the same time, the scientists present— including myself — wanted to know what materials were available and how the experts envisioned how they could use them to complete the symphony.

It's not easy to realize an idea

The task at hand was finally clear. We need to use all of Beethoven's notes and completed works—as well as surviving drafts of the Tenth Symphony—to realize Beethoven's vision.

This is a huge challenge. We don't have a machine that feeds it in a draft, presses a button, and makes it spit out a symphony. Most AI at the time could only add a few seconds of music to an unfinished piece.

We need to push AI beyond its boundaries to do more work. The approach is to let the machine learn about Beethoven's creative process—how he wrote musical bars and painstakingly developed them into exciting symphonies, quartets, and sonatas.

As the project progressed, progress continued to be made on the human and machine parts of the collaboration. Verzova, Gotham, Levin, and Ledell deciphered and transcribed the manuscript of the Tenth Symphony. They tried to understand Beethoven's intentions. Using Beethoven's completed symphony as a template, they tried to solve the problem of which pieces of the manuscript should be arranged in which movement and which part of the movement.

The AI component of the project deals with a range of challenging tasks.

First and foremost, we need to figure out how to adopt a short phrase, or even just a musical theme, as Beethoven has already done, and use it to develop a longer, more complex musical structure. For example, the machine had to learn how Beethoven built the Fifth Symphony on the basis of a musical master title containing four notes.

Second, since the development of phrases also needs to follow certain musical forms—whether it's a harmonic, trio, or fugue—AI needs to learn how Beethoven developed those forms.

Finally, once we have the complete work, the AI has to figure out how to arrange it into an orchestral piece, which involves assigning different instruments to different parts.

And it had to do those tasks like Beethoven did.

Artificial intelligence power

In November 2019, the team met again offline. This time it was at the Beethoven House Museum in Bonn, where Beethoven was born and raised.

The conference offered a test to determine whether AI could complete the project. We printed the score, which was developed by artificial intelligence based on a draft of Beethoven's Tenth Symphony. A pianist performs in a small concert hall at the museum for a group of journalists, musicologists and Beethoven experts.

We let the audience judge which are beetles created by Beethoven and which are the extensions of artificial intelligence to Beethoven's creations. They can't do it.

A few days later, at a press conference, a quartet string band played some of the pieces that ai-powered helped generate. Only those who are well aware of the beethoven Tenth Symphony manuscript can determine which parts were artificially generated.

The success of these trials tells us that we are on the right track. But these are just minutes of music, and there's still a lot of work to be done.

Every moment, Beethoven's genius bursts out and demands that we do better. As the project progresses, so does AI. Over the next 18 months, we constructed two complete movements and organized them into orchestral pieces. Each movement is more than 20 minutes long.

We expect some negative responses to this work. One would argue that AI should not be involved in art, and that AI has no right to replicate the human creative process.

However, when it comes to art, I don't think AI will replace humans, it's a tool that will open the door for artists to express themselves in new ways.

This project would not have been possible without the expertise of human historians and musicians. We do a lot of work to achieve our goals, and of course we have to think creatively.

At one point, one of the music experts on the team said he felt that AI was like a studious music student who practiced, studied, and got better and better every day.

Now, the student has taken the baton from Beethoven and is ready to present the Tenth Symphony to the world.

Source: Reference News Network