Reproduce “Learning to Generate Music with BachProp” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Florian Colombo, Johanni Brea, Wulfram Gerstner
Researchers from EPFL developed BachProp, an algorithmic composer that uses a novel (time-shift, duration, pitch) representation and a deep GRU network to generate music across diverse styles. This approach achieved high prediction accuracies, for instance 97% for time-shifts on Bach Chorales, and offers new quantitative metrics for evaluating generative music models, enhancing general-purpose music composition.
Key insights
- The (dT, T, P) note representation effectively captures rhythmic nuances and polyphony without fixed structural biases, enabling a more style-agnostic approach to music generation.
- Hierarchical prediction of note features within a deep GRU network, combined with shared hidden states, significantly improves the modeling of complex interdependencies between musical parameters.
- Novelty profiles and local statistics provide a robust framework for quantitatively evaluating generative music models, allowing for meaningful comparisons even when different underlying representations are involved.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.