An HCQT variant of BTC for musical chord recognition, plus a reproducible mir_eval benchmark. Honest result: it ties baseline BTC on held-out public data (GuitarSet, Schubert) — not better. Ships the model + weights, the benchmark harness, honest negative results, and a guide to extend the HCQT base to melody/bass/transcription.
-
Updated
Jun 14, 2026 - Python