Harmony beyond twelve tones
ANIMA develops an AI-driven music generation system that focuses on human-centred creativity and the exploration of microtonal harmonies in 31-tone equal temperament (31-TET) and 53-TET. The project establishes a music corpus and computational models that combine novel harmonic trajectories with new chord qualities, applying machine learning techniques to microtonal modes. The goal is to allow diverse cultural approaches while providing composers with new tools for harmonic exploration. The ANIMA project originated as part of the research conducted at the MUSAiC project (ERC-2019-COG No. 864189), directed by Bob L. T. Sturm, in the Royal Institute of Technology (KTH), Stockholm, Sweden.
The Harmonic Eigenspace
The Harmonic Eigenspace is a psychoacoustic model based on Sethares's roughness algorithm that maps the perceived dissonance of tetrad chords (root, third, fifth, seventh). By fixing the root as a reference, the 4D chord space reduces to a navigable 3D topology of consonance and dissonance across microtonal tunings, revealing a landscape of harmonic possibilities.
Generative Models
A GPT-2-based language model trained on microtonal harmony, using a hierarchical tokenization strategy for 53-TET MPE MIDI with a vocabulary designed to capture both vertical and horizontal harmonic structure.
