Evidence-led piano MIDI comparison
How to compare piano audio-to-MIDI accuracy
The most useful converter is not the one that simply outputs more notes. A fair comparison uses the same piano recording, a synchronized reference MIDI when available, fixed settings, and measures that match real DAW editing work.
Guests can try 5 free minutes each month without signing in. Results vary with the recording.
Measure correct, extra, and missing notes
Onset precision measures how many output notes match the reference; recall measures how many reference notes were recovered. Report both and their F1 balance, because a tool can raise recall by adding ghost notes.
Check timing and note duration
Compare onset error and note-off error, including medians and P95. A pitch can be correct while starting late or lasting too long, which creates extra cleanup in a piano roll.
Inspect velocity and CC64 sustain
For matched notes, compare velocity error and correlation. Separately compare CC64 pedal-down intervals and boundary timing. Counting velocity values or pedal events alone does not prove they are correct.
Use the same files and reveal weak cases
Give every tool the same authorized WAV, record the version and default or recommended settings, prohibit manual note replacement, and publish the median, spread, and weakest examples. Without a synchronized reference MIDI, use blinded listening and editing effort rather than an accuracy percentage.
Download a fair comparison worksheet
Use one row per converter and recording. Record the version and settings, leave reference-only metrics blank when no synchronized MIDI exists, and keep editing time plus blinded listening notes separate from objective accuracy.
Download the CSV worksheetCommon questions
Which piano audio-to-MIDI converter is the most accurate?
That depends on a reproducible comparison across notes, timing, velocity, sustain, and editing effort. A single hand-picked recording cannot establish a market-wide winner.
Does a MIDI file with more notes mean better accuracy?
No. Extra notes may be correct recoveries or ghost notes. Precision and recall against a synchronized reference MIDI are needed to tell the difference.
Can a normal room recording provide an objective accuracy percentage?
Not by itself. Without a synchronized sensor or digital-piano MIDI reference, it can support blinded listening and editing comparisons, not objective note accuracy.
Should I convert audio if my digital piano already recorded MIDI?
Use the original direct MIDI when it exists. Audio transcription is most useful for acoustic piano recordings, old audio, phone recordings, or videos that have no original MIDI.