Can a computer create melodies

AI contest: are computers the better composers?

Status: 04/16/20 5:08 p.m.

Karen van Dijk likes to think about the future. No sooner had Duncan Laurence won the Eurovision Song Contest 2019 in Tel Aviv than the editor of the Dutch public broadcaster VPRO was already considering whether it would be possible to repeat this success next year. Maybe there is such a thing as a formula for the perfect ESC song? And maybe artificial intelligence (AI) can help find them? The idea for the first AI Song Contest (AI for "artificial intelligence") was born. A competition that aims to turn data and algorithms into music that would have a chance of winning the Eurovision Song Contest. And since this song contest is carried out online, it will take place unchanged despite the corona pandemic and major events that have been canceled worldwide.

Music at the push of a button? It does not work!

The 31-year-old content producer from Utrecht is not the first to come up with the idea of ​​generating new music from data from ESC songs. Shortly before the Dutch victory, an Israeli collective of musicians and programmers, with the support of song contest legend Izhar Cohen, published the song "Blue Jeans And Bloody Tears". However, anyone who now believes that music can be generated at the push of a button is wrong: "You can use AI to develop melody sequences or musical ideas," explains Karen van Dijk, "but it always takes people to turn them into a real song" . And they have to do a lot of preparatory work before the computer spits out the first melody.

Artificial intelligence provides uinfinitely many results

In fact, any artificial intelligence is only as good as the data it is fed with. All participants were therefore provided with a data set with information on the genre, BPM, key and text of around 200 ESC songs, but far more data is required in order to initially teach the AI ​​how the musical components can be meaningfully combined. Only then can melodies, harmonies, bass lines and texts be created. "In principle, AI can provide an infinite number of results. So at a certain point you have to stop it and choose what you want to use," explains Karen van Dijk. The machine doesn't know how intro, verse, bridge and chorus are related and how they have to be arranged.

German team of scientists with a radical approach

There is still no solution for completely automated singing either. André Röhrig, who together with Ullika Scholz forms one of 13 teams participating in the AI ​​Song Contest, had to pick up the microphone himself: "I put a vocoder over it so that it sounds somewhat harmonious, otherwise the whole thing would have become completely inaudible," laughs the data scientist who otherwise has little to do with music and actually works in the field of image processing. "We just wanted to see what you can achieve with AI when you start programming in order to generate music instead of images or texts." The contribution "Offshore In Deep Water" is therefore perhaps the competition song with the most radical approach when it comes to the use of artificial intelligence for musical purposes.

Expert jury and audience jury vote

At least with the expert jury, Röhrig expects very good chances, because it is not based on aesthetic criteria, but assesses to what extent and how creatively the AI ​​has been used. The audience is responsible for the artistic aspects and can vote here in four categories on the creations of artificial intelligence until May 10th. And even if the contribution of his team was generated by the AI, Röhrig is proud of his musical creation: "If a craftsman uses a hammer, he doesn't say that the hammer made the table." After all, AI is just a tool that is programmed and operated by humans to create music.

AI has long been used by music companies

Do authors now have to fear for their existence? On the contrary, says Karen van Dijk: "AI can inspire people and produce exciting musical ideas that they might not have come up with on their own. It might even create new genres, as was the case with the drum computer and the E -Guitar was the case, "she enthuses. And André Röhrig reveals: "The big music companies already have teams of data scientists who analyze data on current trends and generate music using technology similar to that used by us. The information from YouTube and Spotify has long been used for this." How much AI is there in the ESC contributions of this year?