An AI Completes Beethoven’s 10th Symphony

Ludwig van Beethoven would have been 250 years old in 2020. On 2 and 3 September 2021, in Lausanne then in Geneva, the Swiss orchestra Nexus celebrated the anniversary with two concerts, a year delay due to a pandemic. In a program featuring Brahms and Rachmaninoff, he performed an unpublished work by a German composer. Frankly the one work he didn’t actually write: His 10I symphonies, of which there are only preliminary excerpts, some of which have been attributed by experts to future symphonies without explicit reference to Beethoven.

Thus the first movement was reconstructed and played in 1988. But Nexus Orchestra started from second base: artificial intelligence tools capable of automatically generating scores from a training database. .

Requested by the Nexus Orchestra’s conductor Guillaume Burney, it was researcher Florian Colombo of the Federal Polytechnic School of Lausanne (EPFL) who led the project. An expert in artificial neural networks and a cellist himself, in “deep learning”, he works on technologies to aid structure. Their work gave rise to the Batchprop project, in which sheet music is produced in a particular style of music and composer.

A concert in Bono on October 9, 2021

In addition, Deutsche Telekom also has a target of 10I Beethoven’s Symphony by Artificial Intelligence. A concert is planned in Bonn on October 9, 2021. And in the same spirit, Chinese telecommunications supplier Huawei had “completed” the 8.I Schubert’s Symphony in 2019. But in these two cases, the algorithms create the music, which is then modified, arranged, and segmented by the actual musicians. “Here, my idea is to see how far we can go in automation”, Summaries Florian Colombo. His technologies, around which he founded the start-up AdaMu, provide written scores for each tool in the works.

See also  Fast Galaxy Burst is doing it again with our Galaxy Catched First Star sending fast

The researchers worked with Beethoven’s learning database made up of 16 string checkpoints. He had already used them for a previous project and reintroduced them, believing that they well represented the master’s style of composition. “It also allowed me to go faster, Florian recognizes Colombo. If I had more time, I would have used the symphony’s score.”

However, this database contains all grade related information. For each instrument, three main criteria are analyzed: melody (i.e. what the first violin plays), rhythm, and harmony. Thereby the neural network is able, starting with fragments of an existing score, to “predict” what the continuum will be, respecting the style of a given composer. As part of this project the baptized algorithm, Beethoven, was presented with passages written by Beethoven for his 10.I symphonies and generated musical ideas with their transcriptions. These fragments were retrieved from the IMSLP online library.

A choice between several “possible sequences” of scores

However the role of humans is nothing less: the algorithm can generate multiple predictions from which to choose. This choice then determines the following predictions, the “probable continuation” of the score. “In the beginning, all norms have the same weight, The researcher explains. But you can force an instrument to play a piece like this, and others are built around it. Depending on the decisions you make, the possibilities change. It all depends on who you start with.” In this case, Florian Colombo began by producing the melody, and the bass came next. But if he had attacked the bass first, the melody might have been different. In any case, the neural network doesn’t follow any rules of music composition: everything is based on training from data.

This project is really only one step. Florian Colombo intended to continue to develop a real instrument for composition, including elements of music theory. With the ambition, eventually, that the musician could interact a little more with the technique.

See also  Is Elon Musk's 'Megaconstations' blurred in the night sky? | Satellite

You May Also Like

About the Author: Abbott Hopkins

Analyst. Amateur problem solver. Wannabe internet expert. Coffee geek. Tv guru. Award-winning communicator. Food nerd.

Leave a Reply

Your email address will not be published. Required fields are marked *