MP3 to MIDI: check the noise tail before keeping the notes.
A noisy ending can create extra note events or hide the point where the melody actually stops. Check the tail separately, then compare the MIDI with a silent control before arranging.
Use a recorded control
The public noise-tail record includes a melody file, a gated tail, MIDI outputs, and a results.json file. It is a separate check from the isolated four-note MP3 fixture.
| Input condition | What to inspect | Why |
|---|---|---|
| Clean melody | Expected pitch and timing events | Provides the source comparison |
| Noisy tail | Events after the melody ends | Shows whether the tail is mistaken for notes |
| Gated tail | Output after reducing the tail | Shows what post-processing changes |
What this does not prove
A noise-tail experiment does not establish full-song accuracy or guarantee that every recording will behave the same way. It only documents the tested tail condition and its generated MIDI output.
After conversion
Trim or gate the tail in the source workflow, rerun the same check, and compare the resulting note boundaries. Keep the original and processed files separate so you can see what changed.
Questions about noisy input
Can background noise become MIDI notes?
It can create unwanted events in some recordings. Check the tail output instead of assuming every event belongs to the melody.
Should I remove the tail before conversion?
Yes, when the tail is not part of the musical phrase. Keep the original file and record the processed version separately.
How do I verify the fix?
Rerun the same detector and compare the event boundaries with the earlier result. Use the public noise-tail record as the documented example.