Recording and Evaluating Your Own Pitch Scores

A practical guide to reading the numbers Hearkle gives you, keeping your own log, and not overreading either.

Short answer: a single game score tells you almost nothing. What is worth having is a short log kept over weeks, and three quantities pulled from it: your threshold (how small a difference you can reliably hear, in cents), your bias (do you land consistently sharp or flat?), and your consistency (how much your answers scatter from day to day). None of these is a verdict on your musicality, and none of them is a hearing test. They are rough, honest snapshots, and they are only as trustworthy as your headphones, your room, and your attention allow.

Why one score means so little

Pitch judgements are noisy. The same listener on the same device can be several cents better or worse on two consecutive days, depending on fatigue, background noise, how loud the tone was, and how much attention was actually in the room. On top of that, people differ enormously to begin with: in a large sample of healthy volunteers, pitch discrimination performance spread across a very wide range, with musical training and age among the factors that moved it (Smith and colleagues, 2017). Your first result places you nowhere useful.

There is a second reason to be sceptical of one number. Any threshold that a game reports comes from an adaptive procedure: the difference gets smaller when you are right and larger when you are wrong, converging on the point where you are correct some fixed proportion of the time. Those transformed up-down staircases are efficient and well understood (Levitt, 1971), but they are also estimates with error bars, and a short run has wide ones. Treat a reported threshold the way you would treat a single blood-pressure reading: informative in a series, misleading alone.

The simplest setup: a notebook and consistent conditions

You do not need software. A note on your phone with the date, the game, and the number is enough. What matters far more than the tool is that the conditions stay the same. Same headphones, same volume, roughly the same time of day, roughly the same room. You are comparing yourself to yourself over weeks, so a stable setup beats an expensive one.

Headphones deserve a word of their own. Pitch lives entirely in sound, and a laptop speaker with a weak low end can flatten precisely the harmonic structure the ear uses to derive pitch. If you switch from earbuds to a good pair of over-ears halfway through a month of logging, you have changed the instrument, not the listener, and your trend line is worthless. Pick one and stay with it.

What to measure

Three quantities carry most of the useful information. The table sets out what each one tells you, where to get it, and how not to over-read it.

A fourth thing worth noting by ear is drift within a session: does your accuracy fall off after the third or fourth round? If it does, your sessions are too long. Attention, not ability, is the limiting factor for most people in the first month, and Heardrift in particular rewards a fresh ear.

Separating perception from production

One distinction repays the effort of keeping it straight: hearing a pitch and producing one are different skills, and a low score can come from either. The cleanest demonstration of this comes from singing research. When people were asked to match a target pitch with a slider rather than with their voice, both musicians and non-musicians were more accurate with the slider, which argues against a purely perceptual explanation of poor singing (Hutchins and Peretz, 2012). In other words, plenty of people who "cannot sing in tune" hear the difference perfectly well and cannot yet map it onto their vocal cords.

This matters for how you read your own log. Hearmber, Hearshort, Hearsort and Hearddle are perception tasks: they ask you to judge, not to make. Hearkle, Hearecho, Heardrift and Heartween ask you to place a pitch, which brings memory and control into it. Hearmory and Hearspan add sequence memory and interval reasoning on top. If your judging scores are solid but your placing scores lag, that is a useful, specific finding, and it points you at holding a note rather than at discrimination drills.

Reading the numbers without overreading them

Be gentle with the interpretation. A week is a small sample, and a month is a modest one. What is informative is a pattern that repeats: a threshold whose weekly median steps down twice in a row, a bias that shows up on both Hearkle and Hearecho, a scatter that visibly narrows. One session can fool you; a habit that shows up again and again is worth acting on.

Note too that "exactly on the target" is not always the goal in music. Singers and string players place notes expressively, and ensemble tuning is a moving negotiation rather than a fixed grid. The point of measuring here is to tell a trained, deliberate placement from an unintended miss, and only your ear can decide which is which. If you want the units to mean something, what cents are explains why a twenty-cent error is small in one context and glaring in another.

The honest limits of a browser score

Here is the caveat that has to be said out loud. A browser game running on a consumer device is not laboratory equipment. Audio output timing, buffering, and processing vary widely between devices, browsers, and operating systems, and studies that have measured this directly find real differences in precision and accuracy across platforms (Anwyl-Irvine and colleagues, 2021). For pitch specifically the frequency of a synthesised tone is usually reliable, but the loudness, the frequency response of your headphones, and any processing between the browser and your ear are not controlled at all.

The practical consequence is simple. A score from Hearkle is a relative trend for one listener on one setup, not a certified measurement of your auditory threshold. Comparing your number with a friend's on different hardware is close to meaningless. Comparing this Tuesday with last Tuesday, on your own headphones, is genuinely useful.

And it is not a hearing test. Hearkle is a practice game, not a clinical instrument. It is not an audiogram, it does not control the stimulus level at your eardrum, and it cannot diagnose hearing loss, amusia, or anything else. Congenital amusia is a real and specific condition affecting roughly one and a half percent of people, and it is diagnosed with validated batteries under controlled conditions, not inferred from a browser game. If you have a genuine concern about your hearing, the right step is an audiologist.

Turning a log into practice

Once you have three or four weeks of numbers, the log starts earning its keep, because it tells you what to practise next rather than merely how you did. A high threshold with tight consistency says: keep going, you are stable and the ceiling has not moved yet. A low threshold with wild scatter says: your ear is capable and your attention is not, so shorten the sessions. A clean perception profile with weak production says: move to the holding and reproducing games. That kind of targeted next step is exactly what formative assessment is for, in the sense Black and Wiliam gave the term: evidence gathered during learning in order to change what you do next, not a grade at the end.

If you want a ready-made structure to hang the log on, the four-week pitch plan gives you one game to anchor each week, which makes the numbers comparable by design.

Related reading

• How small a pitch difference can you hear: what a threshold in cents actually means.
• Assessment rubrics for pitch skills: the same idea in observable, classroom terms.
• A four-week pitch plan: a structure that makes your log comparable week to week.

Sources: Smith, Bartholomew, Burnham, Tillmann & Cirulli (2017), Factors affecting pitch discrimination performance in a cohort of extensively phenotyped healthy volunteers, Scientific Reports 7:16480; Levitt (1971), Transformed up-down methods in psychoacoustics, Journal of the Acoustical Society of America 49(2); Hutchins & Peretz (2012), A frog in your throat or in your ear? Searching for the causes of poor singing, Journal of Experimental Psychology: General 141(1); Anwyl-Irvine, Dalmaijer, Hodges & Evershed (2021), Realistic precision and accuracy of online experiment platforms, web browsers, and devices, Behavior Research Methods 53; Peretz & Vuvan (2017), Prevalence of congenital amusia, European Journal of Human Genetics 25; Black & Wiliam (1998), Assessment and classroom learning, Assessment in Education 5(1).