Accuracy · what actually goes wrong
Are vocal range apps accurate?
As accurate as the notes you hand them, which is a different thing from as accurate as the maths. Reading pitch off a microphone is settled: a steady note is measured to within a few cents, finer than anyone sings, and a phone microphone is more than good enough. The error is almost never the detector. It is what the app is willing to count — a breath entered as a top note, a creak entered as a bottom one, a harmonic locked instead of the fundamental. That is why a note here is entered only after 2.0 s of steady pitch inside C2–C6, and why anything thrown out is named on your result rather than quietly folded into the total.
The table below is the five ways a reading goes wrong and what happens to each one here. The scale it is all measured on is 49 notes wide.
The five ways a reading goes wrong
Every one of these is a real complaint about a real tool, not a hypothetical. None of them is a bug in the pitch maths.
| What goes wrong | What it looks like | What happens here |
|---|---|---|
| A breath, or the room | A top note two octaves above anything you sang | Air is not periodic, so it never holds a pitch long enough to be entered |
| A chair creak, or a table knock | A bottom note nobody in the room produced | Same rule at the other end: a knock is over before the hold starts |
| An octave error | A note exactly twelve semitones off, usually the top one | Periodicity in the time domain rather than picking the loudest peak, plus a clamp on both ends |
| A squeak at the top | A range that grows every time you retest | Touched notes are kept and drawn separately from held ones, never added to the total |
| A grunt at the bottom | A low note you cannot reproduce on purpose | A note has to stay steady for the whole hold, so a one-off drop does not survive it |