The Last Moment Before You Fly
From the initial feeling of chaos, from the first impression of nothingness, are born improbable lines that scatter in the surrounding space. Colliding before our confused eyes, the pattern turns around and returns to its core. The meanders – yet linear – embodies the uncertainty where the sounds unfold. Where our mind wanders. Where our thoughts fly away.
The Last Moment Before You Fly
🥉 Third place in the AI Song Contest 2O21
Our song is a co-creative experiment in shaping music, transforming our way of thinking and generating sound. The composer brings his own chords and melodies, which are mixed and re-generated using algorithms designed by the music research team. Seeking to capture new emotions in that AI music content, the composer selects, designs, and organizes chords, melodies, and sounds in subtle lines scattered in the music space.
The contest called for researchers and artists to produce a song through co-creativity and the use of artificial intelligence. We employed a co-creative approach in which the AI and the human fuel each other’s creativity in an iterative process.
We finished in 3rd place among the 38 competing teams. Mark Simos, member of the jury, songwriter and professor at Berklee College of Music, said in the jury session: “What was really unique about that project is the way that the composer got a chance to train the data on their specific individual artistic work, so they felt like they were actually having a kind of conversation with their own sort of stylistic corpus, which gave it a really kind of personal feel." Congratulations to M.O.G.I.I.7.E.D. for their winning song Listen To Your Body Choir, and to all the other teams!
A journey in AI-human co-creation
Generating structured music, involving long-term correlations between elements, is a key challenge in music generation. The approach behind the ambient electronica track The last moment before you fly is structure-based. Through rules and probabilities discussed with the composer, the Algomus team generated song structures, and Sebastien Gulluni selected one, transformed it, and used it as a template for the final track. The structure is both the section layout – here A/B/C/A/CC/D/B/CC – but also indications of global timbre, rhythm, and melody variations such as “rhythm densification” or “ornamentation”.
The composer provided a corpus with his own chord progressions and melodies in other songs. Starting from this music content, candidates for new chords and melodies were generated for each section with the factor oracle, a representation of music memory. Each sequence of two or three chords is found somewhere in this corpus, but the chain of these sequences is unusual, such as the bright transition to Gb Lydian at the half of the song. While he selected the final chord sequences and melodies, Sebastien felt that “It sounds like it has been generated for me!". He arranged and produced the track with a minimalist style, taking great care on the design of each sound.
This co-creative interaction between the AI, the computer music team, and the musician helped to think outside the box. The last moment before you fly is an artifact from the interaction between the composer and these algorithms. This page is regularly updated with more details on the song and the AI/human composition process.
About the team
Sebastien Gulluni is a French composer, sound designer, and music technologist based in Paris. He creates music for media and shapes sounds for companies around the world. His solo project Oblik Lines explores Ambient Electronica territories in both a melodic and textural approach. His compositional process is inspired by his desire to explore new technologies and generative algorithms. Sebastien worked as a research fellow in the music research group GRM and obtained a PhD in Music Information Retrieval.
For The last moment before you fly, Sebastien met the
Algomus research team in Computer Music.
The team works on computer music analysis and notably harmony, rhythm, texture, and music form, in the
CRIStAL lab (Université de Lille), with collaboration from the LITIS lab (Université de Rouen Normandie).
Ken Déguernel, Mathieu Giraud, and Richard Groult model music structure and design analytical and generative algorithms.
With Emmanuel Leguy, they work on the Dezrann platform devoted to create and share interactive music analyses.
We also thank Marie Vautravers, Mariette Giraud, and
Pauline Leroy for their help on concept design and communication.