Return to Aldebaran by Helen Bledsoe Electro Stuff

This is a fantasy sound-design piece worked from an amazing session with tubist Melvyn Poore. Our work was the culmination of my stipendium from the Deutscher Musikrat (2022) to study AI and neural networks in music.
The fantasy description: astronaut Fuzz Baldwin has been kidnapped from Aldebaran B and whisked to the underworld. According to some sources, he was kidnapped along with his pet goose. According to others, he was kidnapped alone but was forced to play the lupophone while in Hell.
Our Heldentubist Melvyn takes it upon himself to attempt an Orpheus-style rescue. Following the sounds of the lupophone (or the pet goose, depending on your source), he sings a hymn to lull the demons into submission. After some discussion about protocol and navigation, Fuzz and Melvyn escape. But the wind has died and they are forced to blow into their sails in order to cross the river Styx. Doomed echos follow them, are the demons waking, or is it the goose snoring?
The technical description: an implementation of the Flucoma tools for Max MSP (developed in Huddersfield, UK) for sound classification to assign different effects to different registers and extended playing techniques of the tuba and the voice. Timbre transfer was generated by RAVE (a Realtime Audio Variational autoEncoder), which was developed at IRCAM by Antoine Caillon and Philippe Esling. The models used for training the encoder were some pre-trained models of NASA recordings, and models that I trained myself on animal sounds and lupophone samples (provided by Peter Veale) with RAVE in Colab.