ABOUT THE TEAM

Spanish team formed by Paula Colmenares León, Gonzalo Nieto Montero and Fernando López Gavilánez respectively, with the additional collaboration of Juan Fernández-Savater Muñoz, as guitarist and guitar arranger.

Paula studies literature, translation and creative writing. Menara is a singer with delusions of grandeur (and her dream of grandeur is not very ambitious, she simply thinks she is famous) who was born from Paula’s navel. Menara allows her to embody different characters and explore the intersection of word and flesh. “I think playing with AI forces you to think in discrete units and how they combine to form a story, and that reminds me of narratology. This experiment will be fun.”

Gonzalo is a musician and sound engineer who currently studies Sound and Music Computing at the Music Technology Group (UPF). Gonzalo is interested in demystifying AI in popular culture, while Gonimix wants to explore the possibilities of new deep learning tools in music creation. Fernando is a telecommunication systems engineer who works on speech processing. He thinks he has found a link between two of his passions: music and science (e.g. Demucs). Since he started using machine learning, he wonders if art can be created tuning weights, losses and optimizers.

ABOUT THE SONG

Coco, duérmete ya is an adult lullaby. Traditionally, the speaker of a lullaby is a mother or a wet nurse who sings the child to sleep. In our song, a mother sings to the coco - a nursery cautionary figure equivalent to the bogeyman - to put it to sleep, afraid that it will cease to exist because the child has already fallen asleep. When no one is afraid of the coco (when the sleeping child stops thinking of it), the coco no longer exists. In the same way, the mother fears that, if she falls asleep and stops thinking of herself, there will be no one to think of her (who will watch over her, who will tuck her in?) and she too will no longer exist. She will sink into the abyss, or into sleep: that which engulfs the coco when the child falls asleep.

ABOUT THE HUMAN-AI CO-CREATION PROCESS

We started the process with a corpus of traditional lullabies in Spanish. With this corpus we fine-tuned (similar to train) a Natural Language Processing model called GPT-2, the Spanish small version. In parallel, we created a melody by continuing (Magenta Studio Continue) and curating another melody of a popular Spanish lullaby known as "A la nanita nana", and reproducing its structure and meter. With the melody and the lullaby generator model, we created the final melody and lyrics. At that point, a collaborating classical guitarist harmonized and arranged the recording that forms the basis of the song (human made). Then, we added three percussion arrangements using Magenta Studio Drumify: one for the kick drum, one for the white-noisy sounds at 0:55 and one for the wooden percussions at the end. In addition to this, we added a pad to support the B-parts using NSynth Ableton Max Patch, plus a sample of Adam Neely's infinite bass solo (generated by Dadabots) in the A-parts. Finally, we mastered the song using IA mastering.