Clever Quest Lab ← Labs

Time Perception Test

Reproduce intervals you see or hear, tap with a metronome and read your pull toward the middle and your scatter, free of any constant device delay.

Settings

Task
Ready and Set are

This run

The results of a run appear here.

History

See all your results

Runs on your device. Uses your microphone; the sound is processed on your device. Nothing you type or record is uploaded. Keeps working without a connection while the page is open. Works with touch. Saves each run's intervals and replies or tap times, your results, whether the microphone was used and your settings, never sound; Clear history deletes runs.

Added · Updated : new Precise mode that times your taps and the clicks in one microphone recording; it no longer fails when the browser drops a few milliseconds of sound, and a double click on Start or Stop does only one thing

What this page asks

Two brief signals mark an interval and you repeat it. Do the short intervals of a set come out too long and the long ones too short? Does your scatter grow in proportion to the interval? Does it matter whether the signals are flashes or clicks, or whether you count in your head? And when you tap along with a metronome, how far ahead of the click do your taps land?

The two tasks

The first time you run a task, a short practice comes first: 6 trials, or 8 taps at 800 ms. It is not counted.

Precise mode (microphone). In the tapping task you can tap the table next to the device in place of pressing a key. The clicks then have to come from loudspeakers, not headphones, because the microphone records them together with your taps. Five clicks at uneven gaps first check that it hears them, so stay quiet for those three seconds. If it does not, you get no number, the run is not saved and you can go on without the microphone. The page asks for the microphone only after you choose this mode and press Start. Nothing you type or record is uploaded: the sound stays in the memory of this tab until each tempo is analysed and is then discarded; only times and results are saved.

Things to try

How the result is calculated

Each trial pairs the interval you got, ts, with your reply, tp. With flashes, ts is what the screen really showed, measured between the frames that drew them, and tp runs from the centre of the Set flash to your press, as in the study of Jazayeri and Shadlen. With clicks, ts is the interval on the audio clock and tp runs from the Set click as it reaches the output to your press. A press before Set or within 100 ms after it is called too early, and the trial comes again later; so does a trial with no reply.

A straight line is fitted through all replies, with weight 1/ts² because scatter grows with the interval. Replies further than 3.5 scaled median absolute deviations from the median of tp/ts are left out, and the result says how many.

tp = a + b·ts     ρ = 1 − b

The regression index ρ is 0 when replies follow the intervals exactly and 1 when you give the same reply to every interval. Its 95% interval is ρ ± t × the standard error of the slope, with t from the t distribution. For each interval i the page takes BIASi, the mean error, and VARi, the variance of the replies, and averages them across intervals:

RMSE² = BIAS² + VAR

Scatter (CV) is the standard deviation of the replies divided by the interval. Dividing by the interval and not by your mean reply keeps it free of a constant delay. If the CV stays the same across intervals, your timing shows the scalar property: scatter in proportion to the interval.

In tapping, each tap is paired with the nearest click within half an interval, and a second tap on the same click is dropped. Asynchrony is tap minus click, so a negative value means before the click. The page reports its mean (uncorrected for device delay), its standard deviation within each run of 30 taps, the standard deviation of the intervals between taps, and the change in mean asynchrony from 500 to 1200 ms. Neighbouring taps depend on each other, so the 95% interval of that change is widened by their correlation from one tap to the next.

Why a constant delay does not matter here. A keyboard, touch screen or sound output adds roughly the same delay to everything. In reproduction that moves a and the mean error, but not b, ρ or the CV. In tapping it moves every mean asynchrony by the same amount, so their change stays. Variation in the delay from one press to the next does not cancel and widens the results.

Precise mode

Sound takes time to leave the loudspeaker and to reach the page through the microphone, and the browser does not always know how long. Here that does not matter: the clicks and your taps are in one recording and only times inside it count, so both delays shift both sounds alike. This is the idea of REPP, the method of Anglada-Tort, Harrison and Jacoby (2022): a free-field recording, with frequency bands to tell the sounds apart. A fingertip on a surface has much of its energy between 80 and 500 Hz, as they report, while the clicks of this page are above 1.8 kHz.

REPP adds special low-pitched markers to its sound and takes the time of every beat from them. This page has simple clicks at known times, so they serve as their own markers: it follows them through the recording and times each tap from the clicks next to it, so that a slow drift between the two sound clocks or a small jump in the recording does not move the later taps. How much the gaps between the found clicks differ from the plan is the timing error in the result. A tap starts where the low band first reaches 40% of the peak of that tap. Sounds no stronger than a click's own trace in the low band do not count as taps; the 4 clicks you only listen to give that trace. The limits (at least half of the clicks found, a timing error up to 4 ms) were chosen for this page. The taps are then paired with the clicks as above.

What studies found

Jazayeri and Shadlen (2010) tested 6 people with this Ready, Set, Go task on an 85 Hz screen. Replies were pulled towards the mean of the set of intervals in use, and they described this with a Bayesian observer who combines a noisy measurement with a learned prior and reports the mean of the result. Scatter that grows with the interval makes longer measurements less reliable, and in that model they lean on the prior more.

Cicchini and colleagues (2012) defined the regression index used here and tested 14 people aged 22 to 33. With flashes, the 6 people without musical training averaged 0.62, while 5 percussionists responded close to the true intervals; with tones, all groups did. In a separate judgement task, at 500 ms with flashes, the mean Weber fraction was 0.15 for the untrained group, 0.11 for string players and 0.06 for percussionists. That task is different, so set these only loosely beside your CV. Their intervals stayed below about 1.2 s, as did those of Jazayeri and Shadlen, so your results for 2 to 10 s have no published mark here.

Taps usually lead the click, and the review by Repp and Su (2013) gives an example at 800 ms: drummers about −20 ms, other groups around −50 ms (a study by Krause and colleagues). Another study they describe, with clicks 1,000 to 3,500 ms apart, found the lead of non-musicians growing with the interval while that of musicians stayed small. In one laboratory's earlier studies, the review notes, the tapping equipment itself added about 15 ms.

Anglada-Tort and colleagues checked REPP against a separate calibration recording on laptops with built-in speakers and microphones: average latency and jitter stayed within 2 ms. Each person's score was the SD of asynchrony averaged over two tasks, tapping to a metronome and to music: measured twice, it gave r = .87 in the laboratory (20 people) and an intraclass correlation of .82 online (166 people). These numbers describe their software and setup, not this page.

Limits

Controls

Sources