The short answer: a hum becomes a single line of pitched notes — pitch, onset time and a loudness-based velocity — but it does not become a faithful copy of how you sang it. Three things a hum carries get left behind. First, vibrato: the MIDI file this converter writes contains no pitch-bend events at all (its writer emits only note-on and note-off messages), so a wobble you sang is flattened to a steady note or lost. Second, slides between pitches become a short staircase of separate notes, not a continuous glide. Third, anything below about 55 Hz — a very low hum — is below the detector's floor and never appears. What survives is useful and editable; it just is not your voice. Every number below came from running this site's own analysis code and MIDI writer on synthesised hums, not from a general description of pitch tracking.
What a hum actually is to the detector
A hum is a single voice: one fundamental plus a few harmonics, without the strong formant that makes a sung vowel misread (the octave-jump case is covered on the voice to MIDI page). Monophonic mode follows it the way it follows any solo line. The ceiling on what comes back is fixed before your take begins: each pitch is estimated once every 11.610 ms, on a search window of 55–1800 Hz. So the question for humming is never "will it understand music" — it is "which parts of a hum fall inside that grid, and which parts does the writer know how to record."
What survives: pitch, a grid-locked onset, and a velocity
In range and reasonably steady, the pitch is right. A 220 Hz hum reads as A3. Note onsets land on the 11.610 ms frame grid, so timing is reproducible to about a hundredth of a second. Velocity is estimated from loudness and capped at 127. That is the entire payload: pitch, time, loudness. There is no timbre, no articulation, no words, no key and no instrument in the export — see what is actually inside a MIDI file and why the file records pitches, not an analyzed key.
Limit 1 — Vibrato becomes a flat note, never a pitch bend
A hummed vibrato is a pitch that wobbles a few cents around a centre. We took a 220 Hz hum and added a sinusoidal wobble at 5.5 Hz, then ran it through this site's own analyser at the default sensitivity and at the top:
| Vibrato depth | Default sensitivity | High sensitivity |
|---|---|---|
| ±15 cents | 1 note — A3 (true pitch) | 2 notes — A3 / A#3 |
| ±25 cents | 0 notes | 1 note — G#3 (misread) |
| ±40 cents | 0 notes | 1 note — G#3 (misread) |
A gentle vibrato (around ±15 cents) comes back essentially as the pitch you sang. A wide one either vanishes at normal sensitivity or collapses, at the top, to a single note that is not even the right one. But the deeper point is structural and does not depend on the wobble's size: the MIDI writer never records a pitch bend. We read the writer's source — track 2 of the file is built only from note-on (0x90) and note-off (0x80) messages, plus one tempo and one time-signature meta event. There is no code path that emits a 0xE0 pitch-bend status. So whatever a hum does, the wobble cannot be in the file; it is either flattened to a steady note or dropped. Some converters do detect pitch bend — this one does not, by construction. The pitch bend detection page lays out exactly what that choice costs and when it matters.
Limit 2 — A slide becomes a staircase, not a glide
Humming loves to slide between notes. We swept a tone from 440 Hz (A4) up to 494 Hz (B4) over half a second and ran it in monophonic mode:
| Input | Notes returned |
|---|---|
| Steady 440 Hz (A4) | 3 notes — A4 / A4 / A4 |
| Glide 440 → 494 Hz | 3 notes — A#4 / B4 / C5, with silent gaps between them |
The glide did not come back as one sliding pitch. It came back as a short staircase of separate notes with silence in the cracks — and because the parabolic pitch estimate overshoots near the ends, the sweep even read past its own top to C5. A portamento you sang is re-quantized into steps. This is the same behaviour a guitar bend shows, which is why guitar to MIDI treats a bend the same way: the writer has no continuous-pitch primitive to put it in.
Limit 3 — Below about 55 Hz, the hum is silent
The detector's floor is ~55 Hz (A1). We tested a set of low hums, monophonically:
| Hum pitch | Default sensitivity | High sensitivity |
|---|---|---|
| 41.2 Hz | 0 notes (below the floor) | 0 notes |
| 55 Hz (A1) | 5–6 notes — A1 | 6 notes — A1 |
| 70 Hz (C#2) | 0 notes | 6 notes — C#2 |
| 82.4 Hz (E2) | 2 notes — E2 | 2 notes — E2 |
| 110 Hz (A2) | 8 notes — A2 | 8 notes — A2 |
Two things follow. The 55 Hz floor is real: a sub-bass hum never appears. But the band just above it, roughly 55–80 Hz, is marginal — at the default sensitivity a 70 Hz hum read as nothing and only appeared when we raised the slider. Above about 80 Hz the reads are solid. There is a second, quieter limit too: below roughly 181 Hz (F#3), one 10.77 Hz FFT analysis bin is wider than a semitone, so low chest-voice hums can read coarse or a note off — the same arithmetic covered on the MP3 to MIDI page. So keep a hum above ~80 Hz if you want clean, confident notes.
Limit 4 — Hum two notes at once and one of them goes
A hum is normally one voice, but if you hum harmony — two pitches overlapping — the modes behave differently. We summed 220 Hz (A3) and 277 Hz (C#4) and ran both modes:
- Monophonic: returned a single pitch (it locked onto C#4 and dropped the A3). A true single line is what mono is for, but by the same rule it cannot hold two at once.
- Polyphonic: returned both — A3 and C#4 — because it looks for several spectral peaks per frame. Overlapping hums can still merge or drop when they crowd the same register, the way any dense material does.
The practical rule: hum one line and use mono. If you deliberately hum harmony, switch to poly and accept that the two voices share one track with no separation — the trade-offs are the same ones on the mono vs poly page.
Limit 5 — The words, the timbre, the feel: none of it travels
The export is a skeleton. The lyrics are not transcribed — the converter reads pitch, not language, so a hummed tune and a hummed word list come back identical. The timbre is gone: a breathy attack becomes a gap in the note, not a soft onset. There is no instrument, no key and no expression written, only pitch, time and a velocity. That is not a defect of humming in particular; it is what every audio-to-MIDI result is, which is why the file is a starting point you then shape — the five passes that turn a raw export into a usable part and the moves that put the feel back.
How to get the cleanest hum-to-MIDI
- Keep the range above ~80 Hz. Below that the reads get marginal or vanish; a comfortable mid-range hum is the safe zone.
- Articulate the notes. Clear breaks between pitches give the segmenter something to hold onto. A slide will still become steps, but clean attacks keep each step the right pitch.
- Record close, not breathy. Breath noise drains the salience score and can drop the note entirely — the mechanism is the same one on the main humming page; a close mic (see microphone to MIDI) and a quiet room (see live vs studio) help more than any slider.
- Use mono for one line, poly only for deliberate harmony. Mono follows a single voice cleanest; poly is the tool for two voices at once.
- Raise sensitivity for low or quiet hums. The 55–80 Hz band only appears at higher sensitivity, as the table above shows — but raising it also lets more noise through, so pair it with a clean recording.
Questions
Will my vibrato show up in the MIDI?
No. The file this converter writes contains no pitch-bend events at all — the writer emits only note-on and note-off messages. A gentle vibrato may survive as a steady note at the right pitch; a wide one is flattened or dropped. The wobble itself is never recorded.
Can it follow me sliding between notes?
Only as steps. A glide resolves into a short staircase of separate notes with gaps, not a continuous slide. If you want clean individual pitches, articulate the slide as distinct notes when you hum.
Why did my low hum disappear?
Either it was below the ~55 Hz floor, or it sat in the marginal 55–80 Hz band where the default sensitivity reads nothing. Raise the sensitivity slider, or hum the phrase an octave or two higher, and it should appear.
Does it write the lyrics?
No. The converter reads pitch, not language. A hummed melody and a hummed sentence come back as the same note list — pitch, time and velocity only.
Should I use mono or poly for humming?
Mono, for a single line — it follows one voice cleanest. Use poly only if you deliberately hum two pitches at once; then it can return both, though they share one track.
Is this the same as other audio-to-MIDI tools?
The pitch tracking here runs on this site's own engine, not a copied one, so the note list is ours. The honest difference worth knowing: this converter does not detect pitch bend or key, so a hum's vibrato and a song's key are not in the file. The accuracy of the notes themselves is covered on the audio to MIDI converter page.
Hum a line into the converter and watch the piano roll as you move the controls — the Audio to MIDI Converter runs entirely in your browser, so every limit above is yours to test in a few seconds.
Related guides
- Humming to MIDIWhat wordless vocals actually come back as
- Voice to MIDIWhat a single-line converter can follow
- Vocal to MIDIWhy a sung note comes back an octave up
- Pitch bend detectionWhat a glide adds to your MIDI, and when it matters
- Microphone to MIDIWhat this converter does with a recording
- Mono vs polyWhat each detection mode misses
- Guitar to MIDIWhat a bend does to your notes
- MIDI converterWhat is actually inside a MIDI file
- Key & scale detectionWhy the file records pitches, not an analyzed key
- Clean up a converted MIDIThe 5 passes that turn a raw export into a usable part