AbilityBenchPitch discrimination
Tone deaf test — can you hear the note that moved?
Two short melodies play in a row and you say whether they were the same; on half the trials one note in the middle of the second has been pushed a semitone out of the key while the shape of the tune is left alone. Free, no signup, and the result is a full breakdown rather than a verdict — sixteen pairs scored for how well the changed ones separated from the unchanged ones and for which way you were leaning, then sixteen more that shrink the change until you lose it and report the smallest one you could still follow, in cents. Nothing here decides anything about you: a weak run through laptop speakers in a room with a fan in it is a measurement of the room.
- 100% free
- No signup
- 32 melody pairs
- Threshold in cents
- Sensitivity and bias split
Two short melodies play, one after the other. Sometimes they are identical and sometimes one note in the middle has moved. 16 trials of that, then 16 more where the change is always there and shrinks until you lose it — about six minutes altogether.
Both blocks, in full
- Stimulus
- two 5-note melodies in a major key, 0.32 s per note, 0.6 s between them
- Block one
- 16 pairs, half identical; the changed ones move one interior note 100 cents out of the key with the shape of the melody kept
- Block two
- 16 pairs, all changed, asking which way the note moved; the size follows a two-down one-up staircase from 100 cents
- Replay
- none, in either block — the comparison is against memory
- Void
- an answer inside the 4.2 s the pair takes to play, or any trial the tab was hidden during
- Floors
- 50% in both blocks, since both offer two answers
- Not the battery
- the Montreal Battery of Evaluation of Amusia is the research instrument; its melodies and its cut scores are its own and neither is reproduced here
Take the practice pairs if you have never done this. Two of the four contain a change and you are told which after each one, so the thing you are listening for is in your ear before anything is scored.
How to take the tone deaf test
Prove a semitone survives your speakers, then judge thirty-two pairs.
Play the two reference pairs before anything is scored
The level check does not just confirm that sound arrives — it plays two tones a semitone apart and then two identical tones, which is exactly the discrimination the first block is built on. If those two pairs sound alike at your current level, raise it; if raising it does not separate them, the honest reading is that something between the browser and your ear is flattening the difference, and six minutes of trials would measure that rather than you.
Judge sixteen pairs with no replay and no feedback
Eight of the sixteen contain a change and eight do not, in an order drawn from the run seed. There is no replay button anywhere in this block, and that is the design: the second melody has to be compared against your memory of the first, which is what makes it a memory-plus-pitch task rather than a spot-the-difference puzzle. Press S for same and D for different, or use the two buttons.
Follow the change down in the second block, then read four numbers
In block two every pair contains a change, so the question becomes which direction the note moved: higher or lower. It starts at the same one-semitone size and halves in on your limit — two right in a row makes it smaller, one wrong makes it larger. The result reports your accuracy on block one against the 50% floor, your sensitivity and your response bias as separate figures, and the smallest change in cents whose direction you could still call.
Technical specifications
| Melodies | five notes drawn from a major scale in a key chosen per trial, 0.32 seconds each, generated in the browser rather than played from a file. The second melody starts 0.6 seconds after the first ends |
|---|---|
| Block one | 16 pairs, half of them identical. On the changed half, one note that is neither first nor last moves ≈ 100 cents out of the key with the up-and-down shape of the melody preserved, so the difference cannot be spotted from the contour alone |
| Block two | 16 pairs, all changed, asking the direction rather than the presence. The size follows a two-down one-up staircase from 100 cents, so it settles at the size you get right 70.7% of the time, and the threshold is the mean of six turning points with its own spread printed beside it |
| Replay | none, in either block. The comparison is with a memory a few seconds old and a replay button would remove the half of the task that is memory |
| Floors | 50% in both blocks. 12 of 16 in block one is where the exact one-sided binomial probability under coin-flipping falls under 5%, which is why raw accuracy is never printed on its own here |
| Bias correction | sensitivity is computed with half a trial added to each cell of the yes-no table, so a flawless sixteen-trial block returns a high finite figure instead of infinity. A visitor who answers different every time scores 50% and a sensitivity of zero, and the two numbers say that plainly |
| Reference figure | Size of an out-of-key pitch change that typical listeners detect and amusic listeners miss is ≈ 100 cents — Hyde & Peretz (2004), Brains that are out of tune but in time, Psychological Science. It is the size at which the published groups separate, not a pass mark, and this page prints your own threshold beside it rather than converting one into the other |
| What this is not | the Montreal Battery of Evaluation of Amusia. That instrument has its own melodies, its own administration and its own cut scores, none of which are reproduced here, so no score on this page is comparable to a score from it |
Frequently asked questions
I sing badly and always have. Is that what this page measures?
No, and the two are much less related than the phrase suggests. This is a perception test: you listen and judge, and never produce a sound. The large majority of people who describe themselves as tone deaf perceive pitch entirely normally and cannot reliably control the muscles that set vocal pitch, which is a motor problem and a trainable one. It is quite possible to score at the top of this page and still be unable to hold a tune, and the page cannot see your singing at all — there is no microphone here, on this page or anywhere on this site.
Why does the changed note keep the same up-and-down shape as the original?
Because otherwise the task can be solved without hearing pitch at all. If the moved note also reverses the direction of a step — a phrase that went up-up-down now going up-down-down — the listener can answer from the shape of the tune, which is a coarser and much easier judgment than hearing that a note is wrong for the key. Holding the contour forces the decision onto the note itself. It is also why the first and last notes never move: an end note has only one neighbor, so half its contour is undefined and there is nothing to preserve.
Why is there no replay button?
Because half of what this trial asks is whether the first melody was still there when the second one arrived. A pitch change inside a melody is not detected by comparing two sounds side by side — it is detected by comparing a sound against a representation you are holding, and that representation decays. Adding a replay would turn a memory-plus-pitch judgment into a pure discrimination task with a different difficulty and a different meaning. The relative pitch test on this site does allow replays, for the opposite reason: there the comparison is the whole task and nothing is being held.
What is a cent, and is a threshold of thirty of them good?
A cent is a hundredth of an equal-tempered semitone, and it is the right unit because pitch is logarithmic while hertz is not — twenty hertz is an octave and a half at the bottom of a bass line and a rounding error at the top of a piccolo. A threshold of thirty cents means you could still tell which way a note had moved when it moved less than a third of a semitone, inside a five-note melody, from memory, through unknown speakers. This page deliberately does not band those numbers into good and bad: it prints your figure, its spread over the turning points that produced it, and the published size at which research groups separate, and lets those sit next to each other.
Why does the second block ask which way it moved instead of whether it changed?
To close the door on a strategy that quietly inflates the first block. On a same-or-different task a listener who is unsure can lean toward different and score well above the floor without hearing anything, which is why block one reports sensitivity and bias as two numbers rather than one percentage. A staircase cannot afford that leak, because it would chase a threshold that does not exist. Forcing a choice between higher and lower removes it: both answers are equally available on every trial, there is nothing to lean toward, and 50% means 50%.
Is this the Montreal Battery?
No. The Montreal Battery of Evaluation of Amusia is the research instrument in this area, with six subtests covering scale, contour, interval, rhythm, meter and memory, and its melodies and thresholds are its own. The battery is the standard research instrument and its cutoffs are defined against its own melodies and its own administration; this file could not verify the threshold values, and they would not apply to a different stimulus set in any case. This page borrows the shape of one of those subtests, generates its own melodies, and reports what you did against chance and against a published deviation size — which is all a browser version is entitled to do.
I did badly. Should I do something about it?
Run it once more on headphones in a quiet room before you conclude anything, because that single change moves scores here more than almost any difference between listeners. If it stays low and it bothers you, the person to talk to is an audiologist or a doctor, and what you would take them is the fact that you noticed, not a number from a website — this page has no diagnostic standing and there is no threshold on it that means anything clinically. If it stays low and it does not bother you, that is also a legitimate place to leave it: plenty of people who cannot hear a semitone move inside a melody enjoy music their whole lives.
What tone deafness means, what amusia is, and why they are not the same phrase
The everyday phrase and the clinical condition point at almost opposite things. Called tone deaf in ordinary speech, most people mean they cannot sing in tune — a production problem, involving the muscles that set vocal pitch and the feedback loop that corrects them, and one that responds to practice. Congenital amusia is a perceptual condition: a lifelong difficulty in perceiving musical pitch, present without hearing loss, without any brain injury and often in people who had ordinary musical exposure as children. The two populations barely overlap. The measurement this page runs is squarely on the perceptual side, which is also why it never asks you to make a sound: there is no microphone path anywhere on this site, by design, and a sung pitch-matching task would need one.
The paradigm comes from the Montreal Battery, whose scale subtest does exactly what block one here does — plays a tonal melody, then plays it again with one note pushed outside the key while the shape is held constant. That manipulation is more demanding than it sounds, and the reason is the contour. A melody carries two kinds of information: the sequence of ups and downs, which almost everybody tracks effortlessly, and the size of each step within a key, which is the part that separates listeners. Move a note far enough to flip the contour and you have asked an easy question; move it enough to break the key while leaving the ups and downs intact and you have asked the hard one. Timing is deliberately not part of this page at all, because the same work that documented impaired melodic pitch perception found timing perception intact in the same listeners — which is why the rhythm test is a separate measurement here rather than a third block, and why folding both into one musicality score would hide the most informative thing about the result.
What browser versions of this test usually get wrong is the scoring rather than the stimulus. A same-or-different task with sixteen trials and no bias correction reports one percentage, and that percentage is compatible with two very different listeners: one who heard eight changes and missed none, and one who answered different on everything and got half of them right by definition. Splitting sensitivity from bias makes those two distinguishable in one line. The second common failure is louder: calling a low score tone deafness. Nothing on this page can distinguish a listener who did not hear the change from one who heard it through a speaker that could not reproduce it, and a page that guesses between those is not measuring, it is labeling. If the ability to hear a note move is what you want measured more finely, the relative pitch test asks you to name the distance rather than notice it, and the auditory processing test covers the layer above both, where sounds are heard perfectly and still do not resolve into meaning.
Congenital amusia prevalence
Share of people with a lifelong deficit in perceiving musical pitch: ≈ 1.5%.
Peretz & Vuvan (2017), Prevalence of congenital amusia · The older and much higher figure: Kalmus & Fry (1980), On tune deafness (dysmelodia): frequency, development, genetics and musical background
Well under the four percent still quoted all over the internet, which came from a self-selected sample decades earlier. It is also not the same thing as being a bad singer: amusia is a perceptual condition and most people who call themselves tone deaf perceive pitch normally and simply cannot control their voice. A page must not convert a low score into this prevalence.
This is a measurement exercise, not a clinical assessment. It reports what you did on this page against a stated reference and nothing more — it cannot establish congenital amusia or any hearing disorder. Only a qualified professional, working with more than a browser, can make that judgment.
Where the thirty-two pairs are generated and scored
Every number on this page is worked out by JavaScript running in the tab you are reading it in. Your answers, your reaction times and your score are never uploaded, logged or kept — which is also why the test carries on working after you disconnect from the network, and why nothing here can be held back behind an email address.
The melodies do not exist until you press start: they are drawn from the run seed and synthesized by oscillators in this tab, so nothing is downloaded and nothing identifies which melodies you heard. Your same-different and up-down answers are counted once, in the page, and are gone with the tab — the staircase history included.