What we tested
We generated two mono WAV files at 22,050 Hz. Each contained four identical 440 Hz A4 tones, each 0.6 seconds long. The only planned difference was the silence between tones: 0.03 seconds in one file and 0.25 seconds in the other. No public recording, video, performer, or channel was used.
Both files went through the local YouTube-to-MIDI converter at its default sensitivity and minimum-note-length controls. The browser finished without console errors and produced one MIDI file for each input.
| Planned gap | MIDI notes | Observed starts | MIDI size |
|---|---|---|---|
| 0.03 s | 4 × A4 | 0.174, 0.824, 1.462, 2.091 s | 86 bytes |
| 0.25 s | 4 × A4 | 0.174, 1.033, 1.893, 2.742 s | 89 bytes |
What the result means
The converter did not merge these repeated pitches into one long A4. It recognized four separate attacks in both files, including the version with only 30 ms of planned silence. The detected starts were close to the generated pulse pattern, while the exported note durations extended slightly across some planned gaps.
That distinction matters in a piano roll: separate notes can be moved, shortened, quantized, or given different velocities independently. One merged note would need to be split before those edits were possible.
Why real repeated notes may behave differently
A generated sine tone has a clean attack and no room sound. A real recording can contain pedal sustain, amplifier decay, compression, echo, another instrument on the same pitch, or a soft re-attack. Those features can hide the moment between two notes. The 30 ms result therefore shows what happened in this test, not a universal threshold.
How to fix merged or split repeats
- Compare the piano roll with the audio. Mark each audible attack before changing note lengths.
- Split a merged note at the attack. Leave a small gap only if that matches the performance.
- Merge false fragments. Join adjacent pieces when there was no real re-attack.
- Adjust velocity after timing. A repeated note often needs its own attack strength once the boundaries are correct.
What this test proves—and what it does not
It confirms four separate A4 notes for two generated inputs with known tone and gap lengths. It does not establish the shortest detectable gap, measure accuracy on public videos, or predict results for piano pedal, guitar decay, vocals, drums, dense mixes, or noisy rooms.
If your draft contains extra fragments, use the extra-note cleanup guide. For long piano tails, see the sustain-pedal test. To inspect loudness changes, read the velocity test.