Gap detection down to 2 ms
Auditory processing test — where hearing stops and listening starts
Three listening tasks that an intact pair of ears can still fail: locating a silence a few milliseconds long inside a tone, reporting three tones in the order they arrived, and holding a contour in each ear when both play at once. Free, no signup, 42 trials in about four minutes, and four separate figures out — a gap threshold in milliseconds, a pattern accuracy against a one-in-six floor, and a score for each ear. They are never added together, because nothing on a website has the standardization sample that a total would need.
- 100% free
- No signup
- 42 trials
- Threshold in ms
- 3 blocks, 4 figures
Three blocks, 18 + 12 + 12 trials, about four minutes. Each one is scored on its own and printed on its own; there is no combined figure at the end, because no browser has the standardization sample that would justify one.
Auditory processing disorder is diagnosed by an audiologist, in a sound-treated booth, through transducers whose output level in decibels is known — because the whole question is what a listener does at a stated level, and neither the level nor the room is knowable from inside a browser tab. What this page can do is show you the shape of those tasks and give you your own numbers on them. What it cannot do is tell you whether anything is wrong, and it will not imply it.
All three blocks, in full
- Gap block
- 18 trials, two 0.7 s bursts at 500/1000/2000 Hz, 1.4 s apart; the silence starts at 20 ms and floors at 2 ms
- Order block
- 12 trials of three 0.5 s tones at 880 or 1122 Hz, 0.3 s apart; 6 sequences possible, all-high and all-low excluded
- Ear block
- 12 trials, three tones per ear at once from overlapping note sets, the ear to report named afterwards
- Every stimulus
- 2.1 s exactly, in all three blocks
- Void
- an answer inside 2.35 s of onset, an answer before onset, or any trial the tab was hidden during
- Floors
- 50% in the gap and ear blocks, 16.7% in the order block
- No composite
- four figures out, no total, no percentile, no cut-off
The warm-up trials run with no clock on the answer and tell you what the right one was. They are marked as practice and are absent from every figure below, including the staircase that produces the gap threshold. The ear block will be skipped until the left-right check above passes: a contour sent to one ear that arrives at both is not a dichotic trial, and scoring it would report your speakers.
How to run the three auditory processing blocks
Two equipment checks, then a staircase, a pattern block and a dichotic block.
Pass both checks — one for level, one for the two channels
The first control plays the same 700 ms burst pair the gap block uses, once with a 20 ms silence in it and once with a 4 ms one, so you hear the range the staircase will walk before you commit four minutes to it. The second plays a tone to one ear and then the other, 0.9 seconds apart in an order drawn at random, and asks which side came first. Getting that wrong keeps the third block out of the run rather than letting it score your speakers, which is the difference between a skipped block and a false result.
Find the silence, then the order, then the ear
Block one gives you two bursts 1.4 seconds apart and one has a silence cut into its middle: say which. Two right in a row makes the silence shorter, one wrong makes it longer, and each burst is presented at a level drawn up to 6 dB down so loudness cannot answer for you. Block two plays three half-second tones at 880 and 1122 Hz and asks for the sequence out of the six possible. Block three sends a rising or falling three-tone contour to each ear simultaneously, from note sets that overlap, and names the ear to report only after both have stopped.
Read four figures and the asymmetry between them
The threshold is the mean of the staircase's turning points with the spread those turning points earned printed beside it, so you can see how much of the number the run actually established. The pattern block is reported against its one-in-six floor with the exact probability of reaching that score by guessing. Each ear gets its own count and its own interval. What is worth attention is not a total — there is none — but a block that came out much lower than the other two.
Technical specifications
| Gap block | 18 trials, two 0.7 s tone bursts 1.4 s apart at 500, 1000 or 2000 Hz drawn per trial. The silence starts at 20 ms, the longest gap the published noise test presents, and follows a two-down one-up geometric staircase to a floor of 2 ms |
|---|---|
| Why 2 ms is the floor | each edge of the silence is ramped over 1 ms to keep a square edge from clicking, and a click is a cue the ear can use instead of the silence. Below about 2 ms the ramps are most of the gap, so the page stops there and says the number is its own limit rather than yours |
| Level rove | each burst is attenuated by an independently drawn 0 to 6 dB. A 2 ms silence removes 0.29% of a 700 ms burst's energy, which is 0.03 dB — the rove buries that difference 200 times over, so the answer cannot come from one burst sounding quieter |
| Order block | 12 trials of three 0.5 s tones with 0.3 s between them, at 880-1,122 Hz — the pair the published pitch-pattern test alternates. All-high and all-low are excluded, leaving 6 sequences and a 16.7% floor |
| Ear block | 12 trials. Left ear takes 440, 587 and 784 Hz; right ear takes 494, 659 and 880 Hz. The two sets interleave, so which side a tone arrived on cannot be worked out from how high it was, and the ear to report is named after the sound has finished rather than before it |
| Every stimulus | 2.10 s exactly, in all three blocks. Three 0.5 s tones with two 0.3 s gaps and a 0.7 s burst pair 1.4 s apart come to the same length on purpose: the anticipation floor is one number for the whole run, so the three blocks have to agree about how long a stimulus is |
| Void trials | an answer inside 2.35 s of onset, an answer before the sound started, or any trial the tab spent in the background. Each is counted under its own reason, replaced, and left out of every figure |
| What is not here | speech in noise, a pure-tone hearing level, a composite score, a percentile and a cut-off. The first needs recorded speech at a calibrated signal-to-noise ratio, the second needs a booth, and the last three need a normative sample this page does not have |
Frequently asked questions
Why is there no speech-in-noise block, when that is the test I was expecting?
Because the number it produces would be a fact about your laptop. A speech reception threshold is a signal-to-noise ratio measured through calibrated transducers at a known presentation level. A browser controls neither the output level nor the headphones, and the visitor's volume slider moves the answer, so the same person produces different thresholds on different laptops. There is a second obstacle that is specific to a web page: the browser's speech synthesizer writes straight to the output device rather than into the audio graph, so a synthesized voice cannot be mixed with a masking noise at a ratio the page controls — and a speech-in-noise task whose ratio drifts with the voice is not a test of anything. The three blocks that are here were chosen because their ratios and intervals are internal to the page: 2 ms of silence is 2 ms of silence whatever your volume slider says.
Does a good gap threshold mean my auditory processing is fine?
No, and the reverse is more useful: a poor one here rarely means anything, while a good one rules out very little. Temporal resolution is one narrow property of the auditory system, measured on one kind of stimulus, and the clinical batteries use four to six subtests precisely because no single one of them is diagnostic. A person who finds 3 ms silences easily may still lose the third word of a sentence in a busy room, because that failure involves working memory, attention and the linguistic prediction that fills gaps in — none of which this page touches.
Do I need headphones, or will speakers do?
The first two blocks work on speakers; the third one does not. A gap in a burst and a pattern of three tones are the same stimulus whichever way they reach you, provided the level is high enough that a few milliseconds of silence is not lost under room noise. The ear block is different in kind: it sends one contour left and another right at the same instant, and a pair of desk speakers delivers both to both ears with a delay and a level difference that varies with how you are sitting. That is why the check gates that block alone, and why passing it on speakers is still a reason to put headphones on.
Why report the sequence of three tones instead of just how many were high?
Because counting is a different and much easier operation than ordering. Two of the six sequences on this page contain two high tones and one low; a listener who only registers the census answers those two with a coin flip. Order is also the part that the clinical version of this task was built around — a pattern is presented and the patient repeats it back — and the reason the task survives normal hearing thresholds is that every tone in it is perfectly audible. Nothing here is quiet or masked. The difficulty is entirely in holding three events in sequence for two seconds.
One of my ears scored much lower than the other. What does that mean here?
With six trials an ear, almost certainly nothing — the two intervals printed beside those counts overlap so heavily that a two-trial difference is noise, and the page says so beside the figures. A real and repeatable ear difference is worth knowing about, and the classic finding is that it depends on what is being played: Kimura (1967) established a right-ear advantage for dichotic speech, while Kimura (1964) found the opposite for melodies, since the two kinds of material are handled by different hemispheres. Tonal contours sit closer to the melody case. If the asymmetry holds up over several runs, an audiologist can measure it with dichotic material that has norms behind it.
Can this tell me whether my child has an auditory processing disorder?
No page can, and a page that offers to is the reason this one says so above the results rather than beneath them. The diagnosis requires a booth, calibrated transducers, an audiologist and a battery administered in a fixed order, partly because several of the subtests are only interpretable once normal hearing thresholds have been established first — a child who mishears because they cannot hear is a different finding with a different remedy. It also depends on age: temporal resolution and pattern reporting both mature through childhood, so an adult reference figure is the wrong yardstick for a young child. What this page can honestly offer a parent is the shape of the tasks, so that an appointment is less strange when it happens.
Why does the page refuse to give me one overall score?
Because a total needs weights, and weights come from a normative sample this page does not have. A clinical battery can combine its subtests because each one was standardized on a known population and the combination rule was validated against a criterion; multiplying a gap threshold in milliseconds by a percentage of patterns and an ear score would require deciding how many milliseconds a pattern is worth, and any answer to that would be invented. The three-figure output is also more useful: the interesting result in this area is nearly always which task fell behind, not how a mixture of them averaged.
What auditory processing means, and what a booth has that a browser does not
The complaint that brings people to this phrase is specific and consistent: hearing is fine, and speech in a busy room is not. A hearing test comes back normal, and the difficulty persists — following a lecture with an air-conditioner running, catching the end of a sentence in a car, telling apart two people who talk at once. What is being described is not detection but everything after it: resolving a sound stream in time, keeping its order, separating what arrived at one ear from what arrived at the other. The three blocks on this page each isolate one of those, which is why they are reported separately, and it is also why the classic batteries are batteries rather than tests.
The gap block measures temporal resolution — how briefly a sound can stop before the interruption stops being detectable. It is worth knowing how small the real numbers are: the reference figure printed beside your result is 4-6 ms, which is between a quarter and a third of the duration of a single frame on a 60 Hz screen. That is the scale at which this system operates, and it is the reason the page can quote a threshold at all while refusing to quote a hearing level. A silence is a ratio of nothing to nothing: it is 3 ms long whatever the volume slider is set to, so the measurement survives an unknown level. A hearing threshold does not, which is why the tone deaf test reports a pitch difference in cents rather than a decibel figure and why frequency-range testing does not live on this site at all.
What browser versions of this get wrong, almost without exception, is the composite. A page that runs three tasks and prints one percentage has silently decided how many milliseconds a mis-ordered pattern is worth, and there is no defensible answer to that question outside a standardization sample. The second common failure is quieter and worse: presenting a dichotic task without checking that the two channels are separate. A rising contour meant for one ear that arrives at both is an easy monaural judgment wearing a dichotic label, and it will produce two high ear scores and a satisfied visitor. The third is the framing — offering a diagnosis, or the shape of one, at the end of six minutes. The rhythm test is the natural next measurement from here, since it asks you to produce intervals rather than judge them, and the Shepard tone illusion is the demonstration that the auditory system is committed to interpretations it cannot be argued out of. If pitch rather than timing is what you are chasing, the perfect pitch test and interval ear training work on named notes and named distances, and the relative pitch test measures the same skill against a given reference. Attention rather than hearing is measured by the Stroop test, which is the other half of what people mean when they say a noisy room is hard.
Gap detection threshold
The shortest silence inside a burst of noise that adults with normal hearing reliably locate: 4-6 ms.
Musiek, Shinn, Jirsa, Bamiou, Baran & Zaidan (2005), GIN (Gaps-In-Noise) test performance in subjects with confirmed CANS involvement, Ear and Hearing
Measured with calibrated equipment, at a stated presentation level, using gaps cut into broadband noise. This page cuts its silences into a tone instead, which is a related but easier judgment, and it cannot know your presentation level at all — so read the two figures as neighbours rather than as the same measurement, and do not treat a millisecond difference either way as meaning anything.
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 an auditory processing disorder. Only a qualified professional, working with more than a browser, can make that judgment.
A central auditory assessment happens in a sound-treated booth, at a presentation level stated in decibels, through transducers with a calibration certificate, and it starts by establishing that hearing thresholds are normal — because a subtest score only means what it is supposed to mean once that is known. A tab in a browser has none of those four things and cannot acquire them. It also has no idea whether you are on a train.
No reaction time is reported anywhere on this page, and that is deliberate rather than an omission: all three blocks are accuracy tasks with a twelve-second window, and answering slowly costs nothing. The only timing that matters here is inside the stimulus, where the audio clock schedules every burst and tone to the sample rather than through a browser timer that would wobble by more than the silences being measured.
Where the bursts, patterns and contours come from
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.
Nothing is downloaded and no microphone is opened — this page asks the browser for no permission of any kind, which is possible because every sound on it is synthesized by oscillators from the run seed. Your answers, the staircase history and the four figures live in this tab only, and reloading the page ends them.