Low-bitrate MP3 to MIDI: what this test shows and what it cannot prove.
A 32 kbps re-encode of the site's four-note source returned the same four detected pitches and timing windows under the same detector settings. That is a result for this fixture, not a full-song accuracy claim.
Test conditions
The source was re-encoded at 32 kbps, decoded to mono 11025 Hz PCM, and passed through the same detector settings: confidence 0.80, minimum note length 100 ms, transpose 0.
| Input | Observed output | Boundary |
|---|---|---|
| 32 kbps MP3 four-note source | Four events: pitches 62, 65, 69, 74; starts 0.1796, 0.8780, 1.5764, 2.2748 s; ends 0.7583, 1.4567, 2.1551, 2.8535 s | Measured on this synthetic melody only |
| 32 kbps silent control | Zero events | Does not test noisy or full-mix audio |
What this does not prove
This run does not prove that low-bitrate MP3 preserves full-song accuracy, transients, vocals, chords, velocity, or multi-instrument separation. It only records the behavior of this four-note source and its silent control.
Reproduce the check
Download the 32 kbps source, the silent control, the run script, and the results record. Compare the MIDI output with the source before arranging.
Questions about low-bitrate input
Did the 32 kbps file lose the four detected notes?
No in this run. The detector returned the same four pitches and timing windows for the tested melody.
Does that mean every low-bitrate song will convert the same way?
No. The test covers one controlled melody, not every song, voice, chord, or noisy recording.
What should I do with a real full-length file?
Run a source-specific check and compare the exported MIDI with the original audio. Treat this page as a reproducible fixture, not a universal quality score.