How the tests work

What each test measures, the cut-offs behind each result, and what none of them can tell you. A cent is a hundredth of a semitone: 100 cents is the step between neighbouring piano keys.

None of these is a medical test

They are quick screens that run in a browser. They don't diagnose congenital amusia (specialists use longer batteries such as the Montreal Battery of Evaluation of Amusia) or hearing loss. If you're worried about your hearing, see an audiologist.

Perfect pitch test

36 notes, each of the 12 note names three times, from C3 to B5, in four synthesized sounds (piano-like, organ-like, triangle and sine). Before every note a second of random tones and noise plays, and consecutive notes are at least a fifth apart. Both stop you naming a note by its distance from the last one, which would be relative pitch, the same idea as Diana Deutsch's online absolute pitch test. There's no feedback during the test.

We report notes named exactly and notes within one semitone. Guessing scores 8% exactly and 25% within a semitone. Our cut-offs: 85% or more exact is a strong perfect-pitch signal; 50% exact or 75% within a semitone shows some absolute pitch; below that, none.

Can adults learn perfect pitch? In one study some adults trained to near-perfect note naming in eight weeks, but they were selected for unusually strong auditory working memory (Van Hedger, Heald and Nusbaum, 2019). We don't promise it.

Tone deaf test

Part one plays two tones and asks whether the second was higher or lower. The first tone roves between 300 and 600 Hz so you can't learn a reference. The gap starts at 200 cents and follows a 2-down/1-up staircase (Levitt, 1971): two right answers halve it (then shrink it by √2 after two reversals), one wrong answer widens it. It stops after 8 reversals or 60 pairs, and your threshold is the geometric mean of the last 6 reversals, the gap you hear about 70.7% of the time. Every sixth pair is an easy 300-cent catch pair that doesn't move the staircase. Missing more than 30% of them marks the result unreliable.

Part two plays 12 pairs of five-note tunes from a major scale. In half, one note after the first moves by one or two semitones.

Our cut-offs: a threshold of 25 cents or less is fine pitch hearing, up to 100 cents is typical. “Not tone deaf” means a threshold of 100 cents or less and at least 60% of tunes told apart. Both above 100 cents and below 60% reads as pitch being hard to hear. Congenital amusia affects about 1.5% of people (Peretz and Vuvan, 2017).

Singing test

You first hold a comfortable note; ten target notes are then taken from the major scale within five semitones of it. Each target plays and stops before you sing, so the microphone hears your voice rather than our speaker. Your pitch is tracked about 60 times a second with the McLeod pitch method, frames below 0.88 clarity are ignored, and brief octave jumps are rejected. A note's error is the median of your held pitch after the first 250 ms (the scoop onto the note), folded to the nearest octave, since singing a note an octave away is singing the right note.

You don't see your pitch while you sing, only afterwards. Seeing it live would let you slide onto the note, and the test would measure that correction rather than your unaided singing. Live feedback is what practice is for.

In tune means within 50 cents. Our cut-offs on the average error: 25 cents or less, very accurate; 50 or less, in tune; 100 or less, a little off; above that, often off. Most occasional singers in one study sang about as accurately as professionals when they sang slowly (Dalla Bella, Giguère and Peretz, 2007).

Vocal range test

Only notes held within 50 cents for at least 0.4 seconds count, so slides, creaks and squeaks don't stretch your range. Voice type is the classical range your range overlaps most, with ties broken by the closest centre: bass E2–E4, baritone A2–A4, tenor C3–C5, alto F3–F5, mezzo-soprano A3–A5, soprano C4–C6. Real voice type also depends on comfort and colour.

Your microphone and your data

Sound from the microphone is analysed in your browser and discarded frame by frame. Nothing is recorded or uploaded. When you finish a test we save its score (for example “18 cents, 92% of tunes”), your device type and a random device id, so we can later show how your result compares with everyone's. See the privacy policy.

Sources