Classical Library

145 public-domain classical recordings from the Internet Archive Musopen collection, streamed from Cloudflare R2. We're building toward a model that takes a reference to an audio file and returns a JSON description (energy, mood, and more). The energy/mood labels you see are deterministic — derived from tempo markings, key, and the known character of famous works — not model output. They're the ground truth we'll judge the model against: run it on a track, compare its JSON to this baseline, and see where it agrees or disagrees.

Caveat: labels are heuristic. Work-level energy is an aggregate (with famous-work adjustments), and movement-level energy is tempo-based. Inside a "Calm" work there can be fiery passages, and vice versa — e.g. the funeral march sits inside the "Energetic" Eroica. Next steps: finer per-movement classification, then the model pass.