Yandex Neural Network Co-authored a play for viola and orchestra

In the last season of The Simpsons, the main character found out: to charm geeks, you need to start a song that a neural network composed. Everyone will be spellbound to listen and dance. Yandex has almost realized this idea.

Composer Kuzma Bodrov said that his last work - Composition for Viola and Orchestra - was not written alone, but in collaboration with the Yandex neural network. Specially trained in classical music, the neural network generated a musical series, on the basis of which Kuzma was able to write a melody and develop it into a whole play. The musical premiere of the opus will take place at the gala concert of the closing of the XII Winter International Art Festival in Sochi.

“The idea was a joint one between Yuri Bashmet and Dmitry Grinchenko,” says the composer. - I received about twenty files with something from the Yandex neural network. These were the melodic lines laid out on the piano. Oddly enough, there were interesting grains in them, for which rumor clings. But they were not framed. This went in a single pointless stream, as is the case with unhealthy people. Although sometimes there were rhythmic formations in this stream. I took several of these topics from there and polished them a bit. Then he continued the more conscious building of the form, did the orchestration and brought out a part of the solo viola. It turned out such a work. "

One cannot expect that a neural network will reliably deliver musical masterpieces. However, it is capable of generating an infinite amount of musical material in which interesting fragments come across. This can help authors who are looking for an idea: “Composers, when they start composing, usually encounter a problem: what to cling to, which thread to pull to find a melodic turn, or chord, or timbre,” Kuzma Bodrov explains. - And this found begins, as it were, to grow. There is always difficulty in this search. And in this case it was easier. These grains have already been set. I got the bricks from which I began to build the building. This can be called co-authorship, because the melodic lines were proposed not by me, but by the neural network. ”

Kuzma Bodrov jokes that the royalty should really be paid to both authors, because this is a full co-authorship.

The author said that the duration of the work is 7-8 minutes. The author pushed himself away from the English “genre” genre, where the melody is repeated, and everything around is changing. At the same time, he decided to choose the fast pace that the Yandex neural network offered him. He gave the proposed melody to the solo piano, began to loop many times and build variations around.

The name for the work has not yet been invented. The play is still listened to by the editors of Yandex and may offer some sensible ideas.

This is not the first musical work of the Yandex neural network. In 2016, she helped compose the album Neural Defensemade in the style of Yegor Letov and the group "Civil Defense". True, there the work of Artificial Intelligence was limited to writing poetry.

"Neuropoet" at "Yandex" has been working for a long time. In 2017, he composed poems in the style of Kurt Cobain , the lead singer of Nirvana. That music project was led by Alexei Tikhonov and Ivan Yamshchikov, an employee of the Max Planck Institute for Scientific Mathematics in Leipzig. At the same time, the authors published a preprint of a scientific paper on the generation of music by a neural network.

Most likely, the new project of automatic music generation is based on the same scientific work. Article Music Generation with Variational Recurrent Autoencoder Supported by HistoryIt was updated several times, and the latest version is dated November 12, 2018. That is, it is clear that the authors continue to work on the project. Variational Recurrent Autoencoder Supported by History (VRASH) scheme for generating music. By clicking on the enlarged image , Yandex will broadcast a concert where they will perform the Composition for viola and orchestra, written in collaboration with the neural network. The broadcast will take place on February 24 at 19:00 (Moscow time). The broadcast can also be launched on the website of the channel "Culture" .