France - Algomus & Friends

If you are looking for a group of musical scholars, better call team Algomus. This five-member research team at the universities in Lille and Amiens are experts in music analysis with computers. Although they all play a musical instrument, they’re not used to generating new music with this academic knowledge, and have little experience with Song Contest entries. Time to put them to the test.

The starting point of Team Algomus is cooperation between humans and AI. ‘Creativity requires a careful balance between all the elements in a larger whole in order to evoke an emotion, and is inextricably linked to humans,’ they say. Nevertheless, they tried to rely as much as possible on AI for the sake of the contest. To this end they mainly used the dataset of Song Contest entries as a basis, but also added models from their own research and the academic community. 

Each section of the song – verse, chorus, and so on – was approached separately to create a chord scheme, lyrics and melody. For the lyrics, the team started from the most common word pairs in Eurovision songs, like 'my heart' and 'the sun', and used this as the basis for the AI to create the lyrics. The first round was an instant hit: a good balance between grammatically correct sentences and room for poetic interpretation. When making the music, a intervention was filtering material generated by AI. This was done by music experience and feeling: which structures, melodies and chords don’t feel right? Then the team left the fate of some decisions to chance. The percussion and bass line were composed by the team, and the chorus melody was formed on the basis of chords generated by AI.

Finally, they visited Niam, an artist and student at the University of Lille, to sing in the lyrics. Her voice is combined with a synthetic one. As music lovers, the team members sometimes found it difficult to let go of the strange lyrics and chords and illogical structure created by the AI. According to them, this is where the core of the collaboration with AI lies: to make a song right, the human plays with the AI through the song writing, and, in the end, perform the final touch.

Three of the team members are connected to the university in Lille. Louis Bigo was familiar with AI musical technology and studied the trance genre. Mathieu Giraud specialises in bioinformatics and harmony, while Gianluca Micchi has a background in physics and is now focused on deep learning. Florence Levé and Richard Groult are in the Université de Picardie Jules Verne. Florence Levé has a background in combinatorics, meaning the mathematics of counting, and Richard Groult in algorithms on texts. The team members jointly work in music information retrieval, a research field that focuses on the understanding and use of music data.