What we tested
We generated a 22,050 Hz mono WAV with four isolated 0.9-second notes: C4, D4, E4, and G4. Each note used a fundamental plus harmonics shaped around broad vowel-like formants, with a gentle amplitude pulse and 0.3 seconds of silence between notes. The source was deterministic synthesis; no human recording or public media was used.
The file went through the local YouTube-to-MIDI converter at its default controls. The browser completed without console errors and exported a 130-byte MIDI file with one musical track.
| Input note | Main MIDI note | Main duration | Additional output |
|---|---|---|---|
| C4 | C4 at 0.244 s | 0.952 s | C5 for 0.070 s and D7 for 0.673 s. |
| D4 | D4 at 1.451 s | 0.942 s | D5 for 0.081 s. |
| E4 | E4 at 2.636 s | 0.964 s | E5 for 0.116 s. |
| G4 | G4 at 3.856 s | 0.942 s | G5 for 0.814 s. |
Why extra notes appeared
A vowel-like sound is richer than a sine wave. It contains the sung fundamental plus stronger frequencies above it. The current transcription found the intended fundamentals, but it also treated several harmonics as separate pitches. In this file, the octave extras were C5, D5, E5, and G5, plus one D7 event.
That makes the result a starting point rather than a finished vocal transcription. The file contained nine note events and zero pitch-bend events. It stored note instructions, not the source words, pronunciation, breath, or vocal timbre.
How to get a cleaner vocal MIDI draft
- Start with one dry voice. Remove accompaniment, doubled vocals, and long reverb when possible.
- Keep the melody exposed. A clear monophonic line gives the detector fewer competing pitches.
- Review octave notes. Delete short or overlapping notes one or more octaves above the melody when they follow the same timing.
- Rebuild expression in the DAW. Add bends, modulation, articulation, and a suitable instrument after cleaning the note line.
What this test proves—and what it does not
It records the current converter's exact output for one synthetic, isolated four-note vowel-like source. It does not measure every singer, lyric, range, microphone, harmony, effect, or finished mix. A real vocal can produce a different note count.
For stacked parts in a full mix, read the instrument-separation boundary. If the draft contains short unwanted notes, use the extra-note cleanup test.