Concordance is a polyphonic music transcription model trained on audio paired with scores that are not time-aligned to it. Its core loop borrows a published method — an expectation-maximization procedure that uses the model's own predictions to warp an unaligned score onto a recording, manufacturing frame-level labels with no manual alignment — and pairs it with a data core of public-domain classical guitar, so the released weights can carry a permissive license rather than the non-commercial terms that encumber the field's existing models.