ABOUT THE TEAM

Our team consisting of colleagues, friends and family all hail from The Netherlands. The team started out with two colleagues and the whole thing just snowballed into a full-fledged team. Our data scientists are: Hennie, played Xylophone and piano for over 10 years and likes to learn from these type of contests; Rick, humanities major and notoriously a-musical; Salomon, jazz and piano player in a big band; Vincent, guitar player and always wondering how things work. And our musicians: Joëlle, singer-songwriter that ditched the whole music part to become just a writer; Thomas, composer and specialized in writing for film and games; Wendel, wellness manager by day, DJ by night (which still has something to do with wellness really). We set out on a dual quest: to find out if AI can become a new band member and to test some nice algorithms we don’t get to use during working hours.

ABOUT THE SONG

Honestly, saying something profound about what people should feel when they hear our song wouldn’t be very honest. Our song is entirely build up by computers. Sure, we decided on a theme, but also went a bit ‘meta’ on that part and let it learn from a subreddit about science-fiction concepts. Almost every piece is either made by AI or influenced by AI. The only purely human activity was the building the song structure and the production. So what is the song about? We guess it’s about letting go of our human notion of what music should be like. Nothing profound, just a bunch of autonomous musicians surrendering to the whimsical wishes of AI. We have no idea what a Quantum Trap is.

ABOUT THE HUMAN-AI CO-CREATION PROCESS

We used language models to generate music. We changed music into a symbolic textual representation. Just like text, music has a pattern of logical succession. Our first model could predict the next chord when giving a chord; this way we generated chord progression. Then we fine-tuned an existing piano model to create chords, melody and bass. To go back full circle we matched melody lines and chords, allowing for conditional melody generation based on chords.

Lyrics were made by feeding an algorithm posts from the SciFiConcepts-subreddit. The output: some freaky scifi ideas, but not so great lyrics. The generated texts were thus put in a different model, which we trained to transfer Wikipedia-like texts into lyrics.

AI is very scalable. So, we ended up with a few hundred songs and lyrics. To find something nice we trained models that used Hot 100 Billboard-data to determine if a song was interesting. To cherry pick the lyrics, we build a mood labeler algorithm for both the music – using visual representation of music – and lyrics. This allowed us to match the lyrics to the music based on their emotional pattern.

Then we did a human-machine jam session. Our producer build the song structure with the music that made the cut – and when he needed variations, we send the songs back to the algorithms. Near real-time creating or adding new parts. Drum beats were made. Singing was recorded. And after some fine-tuning and mixing, we ended up with our final song.