Piano to MIDI

Scores come from the linked base checkpoint, not a new measurement of the model running here.

Public sources checked: 2026-09-12

Scores come from the linked base checkpoint, not a new measurement of the model running here.

Public sources checked: 2026-09-12

Scores come from the linked base checkpoint, not a new measurement of the model running here.

Public sources checked: 2026-09-12

Verified public/reproducible MIREX rows compare average note-onset F1 across three datasets and holistic note F1 across Maestro and SMD.

Public sources checked: 2026-09-12

Official MIREX MAESTRO results, including all published note and pedal P / R / F1 (%); unreported metrics remain absent.

Public sources checked: 2026-09-12

Official MIREX MAPS results, including all published note and pedal P / R / F1 (%); unreported metrics remain absent. MAPS uses ad hoc alignment; interpret offset metrics in that alignment context.

Public sources checked: 2026-09-12

Official MIREX SMD results, including all published note and pedal P / R / F1 (%); unreported metrics remain absent.

Public sources checked: 2026-09-12

MAESTRO and MAPS results from Table 1 of the Aria-AMT paper, compared separately by dataset. See the model notes for the relationship between paper results and released weights.

Public sources checked: 2026-09-10

Values come from the linked public source. Row labels identify the test set or checkpoint; models without a matching result are omitted.

Public sources checked: 2026-09-10

MAESTRO note results from Table I and sustain-pedal results from Table VI, including published frame, onset, offset and velocity P / R / F1 (%). Unreported metrics remain absent.

Public sources checked: 2026-09-10