Creating new Intermate hit songs using AI

Within intermate there exists a very specific and sometimes strange culture. Together with this culture also a very specific sub genre of music is intertwined. If you are curious about the most popular music within Intermate you can read the following article: The songs you must know as a new Intermate member!

For an outsider it is hard to grasp why this music is so popular within Intermate and it is even harder to find new music that could fit within this genre. Sometimes new songs will be introduced but this is always very inconsistent and you never know if something will stick. To make this process of finding new Intermate songs a bit easier and more consistent I got an idea. What if an intelligent algorithm will create new Intermate songs for us that immediately fit the genre we already have. New fresh and hip songs → fun party’s → happy Intermate members → productive learning hours → higher grades → better jobs when we graduate → more money. So having more songs will lead to more money, who doesn’t want that? So to make this possible I went on a journey to create new songs using Artificial Intelligence. 

After reading thousands of research papers I stumbled on a new technology called jukebox by OpenAI. Using this technology I would be able to analyse already existing songs and generate new variations of these songs. To test out the potential of this AI algorithm I picked three Intermate hits as a basis to create new songs from. These hits were; ‘If I tell you’ by René le Blanc, ‘Maren-Kessel Tunnelrave’ by MC Ludo and Qriminal and a new song on the block: ‘Als jij je vinger in mn kont steekt’ by Ron van Zalmsaus. Using the first 12 seconds of each song the algorithm generated what it thought the rest of the song would sound like, which resulted in a completely new song. 

The first song that the algorithm generated, which was based on ‘If I tell you’ by René le Blanc is the following masterpiece. https://on.soundcloud.com/HTd6k

There is still some noise present in the song because it would take more than 12 hours to generate a fully denoised song. I did not have this time and resources so I chose quantity over quality. As you can hear, the rich voice of Rene can be heard throughout the song and the generated piece captures the music genre perfectly. If you ask me, this song is an instant classic. 

The next song that was generated was based on ‘Maren-Kessel Tunnelrave’ by MC Ludo and Qriminal. https://on.soundcloud.com/MdSrg. 

As can be heard the AI generated song chose to stay silent sometimes in contradiction to the original song. In my opinion this is a very creative lyrical choice and gives a new meaning to the song. I think this song is more for the people who want to listen to Maren-Kessel Tunnelrave on a lazy Sunday afternoon. It is a more chilled out version of the original. 

The last song that the AI algorithm created was based on ‘Als jij je vinger in mn kont steekt’ by Ron van Zalmsaus. https://on.soundcloud.com/QfWuX

In this song you hear a new voice coming in that is different from Ron van Zalmsaus. I think this new voice blends in perfectly with the very distinct voice from Ron. You hear the music building up creating a new rhythm and beat that makes it a bit different from the other songs while still keeping that van Zalmsaus vibe. Again an instant classic I would say. 

It is scary how well these songs capture the Intermate style we were looking for and I would call this experiment a massive success. They might still need a bit of tweaking and especially denoising but when given enough time I am certain they will finish 1, 2 and 3 on the Intermate song billboards, if not higher. 

Are you interested in creating AI generated songs for yourself, or improving the songs that I have already made? The codebase I used to generate these songs is open source and can be found here: https://github.com/openai/jukebox/