MP3 to MIDI: a silence gate can decide whether a quiet note appears.
In this controlled amplitude run, the detector returned no note at peak amplitude 0.007 and returned one note at 0.008 with confidence 0.8. This is a synthetic threshold observation, not a recording-level accuracy guarantee.
What was tested
The exact deployed detector was evaluated on one 1.4-second synthetic A4-like event from 0.2 to 1.2 seconds. The run used mono 11025 Hz PCM, confidence 0.8, minimum note length 100 ms, and transpose 0. Full records are in results.json.
| Peak amplitude | Observed output | Boundary |
|---|---|---|
| 0.006 | No clear notes found | Below this run's observed threshold |
| 0.007 | No clear notes found | Still below this run's observed threshold |
| 0.008 | One detected note, pitch 69 | Observed only for this synthetic event |
| 0.01 and above | One detected note, pitch 69 | Does not test real noise or a full mix |
What this does not prove
This run does not prove a universal silence gate, a minimum recording level, or full-song accuracy. Browser decoding, room noise, compression artifacts, and mixed instruments can change the result.
How to use the observation
If a real melody returns no notes, first check whether the source is extremely quiet and isolated. Lowering confidence may change the output, but that tradeoff must be checked against the original recording rather than assumed from this synthetic threshold.
Questions about quiet input
What happened below amplitude 0.008?
The tested 0.006 and 0.007 inputs returned no clear notes at confidence 0.8.
Does 0.008 mean every real recording will work?
No. It is the observed transition in one synthetic event under one detector configuration.
What should I keep with the result?
Keep the source, detector settings, and results record together. That is what makes the threshold observation reproducible.