MP3 to MIDI: reproduce one measured melody check before arranging.
Use the same source, decode settings, detector settings, and output comparison every time. This checklist turns the site's existing four-note MP3 fixture into a repeatable handoff instead of an accuracy promise.
Start with the exact input
Use the public source.mp3 fixture and keep the detector input at mono 11025 Hz PCM. The melody is the site's four-note seed; do not substitute a different recording when comparing runs.
Run the same detector
Use the published app.js and the recorded run script. Keep confidence at 0.80, minimum note length at 100 ms, and transpose at 0 so the output remains comparable to the source record.
Check the output, not just the download
Open the generated MIDI and compare event count, pitch sequence, and each note's start and end window with results.json. Also run the silent control; a silent input should not be treated as a successful melody conversion.
| Check | What to record | Why it matters |
|---|---|---|
| Input | File name and decode target | Prevents comparing different source material |
| Notes | Events, pitches, starts, ends | Shows what the detector actually returned |
| Control | Silent-control result | Separates a real melody result from an empty run |
After the check
Keep the MIDI as editable note data, then compare it with the source before arranging. If the passage contains chords, drums, clipping, or several instruments, record that as an untested condition rather than extending this isolated-melody result.
Questions from the checklist
Does a downloaded MIDI prove an accurate MP3 to MIDI conversion?
No. It proves that a MIDI file was produced; the measured event and timing comparison still has to be checked against the source.
Why include a silent control?
It shows whether the same detector run returns no note events for silence. That keeps an empty input from being mistaken for a useful melody result.
Where can I inspect the original records?
Use the linked source file, script, detector, and results JSON. They are the public evidence chain for this checklist.