Singing, measured

Across 632 measured voices the median range is 23 semitones — just under two octaves — running D3–C#5. Half of all singers land between 17 and 30 semitones, so a range under two octaves is the ordinary case rather than a shortfall.

The number people expect is usually larger, because published ranges belong to professional singers who trained for years and are quoted at their extremes. A 23-semitone working range is enough to sing most popular music, provided the song sits in the right octave. Measure your own range and you can place yourself in this distribution directly.

The vocal line of the median uploaded song spans 20 semitones, with the middle half between 16 and 25. Set that against the median measured voice at 23 semitones and the gap is only 3 semitones: a typical song nearly fills a typical range, leaving almost no margin for singing it in the wrong octave.

That is why “can I sing this?” is usually a range question before it is a technique question, and why the answer changes when the song is shifted by an octave. 26% of these songs (87 of 336) span more than two octaves, which no octave shift can rescue for a median voice. Per-song range data is published for each song we cover on the song database; how each figure is computed is described in the song-page methodology.

Pairing the first and best scored attempt each time a singer repeated the same piece — a full-song run or a single drill from a song’s journey — gives 604 pairs from 171 singers. The median relative gain is 39%, 77% of pairs improved at all, and the median best score landed on attempt 2.

Read this narrowly. It measures improvement on the song that was practiced — a learning effect on one piece of music, not evidence that general singing skill transferred to anything else. It also says nothing about singers who tried a song once and never returned, since a single attempt cannot be paired. Across the same population there are 3961 scored attempts by 263 singers overall.

Vocal ranges come from the app’s guided voice exploration: the singer slides up and down on an open vowel while on-device pitch detection records the lowest and highest note held steadily. One profile per person, the most recent measurement. Profiles missing an end, or with a zero or negative span, are dropped.

Song spans are measured from the vocal line the audio pipeline extracts from each uploaded song — the lowest and highest note of the melody itself, not of the full mix. Difficulty, tessitura and phrasing tempo are computed the same way; the formulas are written out in the song-page methodology. Auto-seeded starter songs and deleted uploads are excluded.

Improvement is the median of the per-singer, per-piece relative gain between the first and best scored attempt on the same song run or journey drill, over pairs with at least two scored attempts. Scores from different drills are never compared to each other. Highway song runs and song-journey drills count; unscored practice does not.

Exclusions. Internal, test and staff accounts are filtered out of every figure by email domain, so the numbers describe real users only.

Limitations. This is a self-selected population — people who chose a singing app — not a random sample, so it should not be read as describing the general public. The range test is self-administered: a singer who pushes gets a wider result than one who stops at comfort, and neither is verified by a teacher. Nothing is normalized by age, gender or training background, all of which move vocal range. And the improvement figure is confounded with familiarity: repeating a song improves the score partly by learning the song, which is exactly what the app is for but is not the same as becoming a better singer in general.

These figures are free to cite, quote and chart, including commercially, under CC BY 4.0. Attribute them to PitchHighway and link back to https://pitchhighway.com/research. Please carry the as-of date (2026-08-05) with the number — the snapshot is refreshed quarterly and the values move.

For the underlying breakdowns, a comment, or anything not answered here, email [email protected]. Logos, screenshots and boilerplate are in the press kit.