"Golden ears" is an audio industry term for someone who reliably hears differences other people miss. It has no certification behind it and no agreed threshold, which is exactly why it gets thrown around so loosely — by mastering engineers who have earned it, and by forum posters who have not.
This test puts a number on one measurable component of it: pitch resolution. How small a frequency difference can you actually detect? The trials run from 20 cents down to 4 cents, where a cent is a hundredth of a semitone. At the bottom end you are being asked to hear a twenty-fifth of the gap between two adjacent piano keys.
Two tones play in sequence. Say whether the second was higher or lower
than the first. Ten trials, narrowing each time: 20, 15, 12, 10, 8, 6, 6, 5, 4, 4 cents.
The base frequency is randomised between 150 Hz and 800 Hz on every trial, so there is
nothing to anchor against but the pair in front of you.
Desktop: press P for higher, L for lower.
Mobile: tap the top half of the screen for higher, the bottom half for lower.
Headphones and a quiet room, or the test measures your speakers rather than your ears.
Cents are the unit that makes pitch differences comparable across the register — 10 cents is the same perceived interval whether you are at 100 Hz or 1000 Hz, because cents describe a ratio rather than a number of hertz. Some reference points, so the scores above mean something:
| Interval | Cents | What it is |
|---|---|---|
| Octave | 1200 | A doubling of frequency. |
| Semitone | 100 | Two adjacent keys on a piano. Failing to hear this is the scale of gap that defines tone deafness. |
| Pythagorean comma | ≈23.5 | The amount twelve stacked pure fifths overshoot seven octaves by. The reason equal temperament exists. |
| Syntonic comma | ≈21.5 | The gap between a pure major third and the one you get from stacking fifths. |
| Tempered major third error | ≈14 | How sharp the major third on a piano sits against a pure one. Audible as beating, and the reason tempered thirds sound restless. |
| Practical discrimination limit | ≈5–10 | Where careful, trained listeners typically stop being reliable in sequential comparison. |
| Professional tuning tolerance | ≈1–2 | What a good piano tuner holds in the mid-register — but with simultaneous tones, where beating gives them a cue this test deliberately removes. |
That last row is the honest caveat on this whole exercise. Tuners and string players hear far below 4 cents when two notes sound together, because interference beats turn a pitch difference into a rhythmic pulse that is much easier to detect. Play the notes one after the other, as this test does, and you strip that cue away — leaving pure pitch memory and comparison, which is a genuinely harder task.
"Golden ears" grew up inside professional audio, where the ability to identify what is wrong with a mix or a loudspeaker is a job skill rather than a party trick. It got formalised into training material — Philips built a Golden Ears program to develop listening skill in engineers, and Dave Moulton's Golden Ears course, originally issued on CD, became a standard for learning to name what you are hearing rather than just feeling that something is off.
The most useful research on it comes from blind loudspeaker evaluation, particularly the work Floyd Toole ran at Harman. The finding that matters: trained listeners reliably detect relatively small frequency response differences that untrained listeners do not, and their judgements agree with each other far more tightly. Training does not give people better ears. It gives them better questions — a vocabulary and a set of expectations that direct attention to the right part of the sound.
Golden ears are real for real differences. They evaporate for imaginary ones. Both halves of that sentence are supported, and leaving either one out is how the topic becomes tribal.
When a difference genuinely exists in the signal — a response deviation, distortion, a level mismatch, a timing error — trained listeners find it, describe it consistently, and outperform novices by a wide margin. When a difference does not exist in the signal, the same listeners often report hearing one anyway, and those reports collapse the moment the test goes double-blind. This is why ABX testing exists, and why sighted listening is treated as evidence of nothing in serious audio work: expectation is not a bias you can concentrate your way out of, and being highly skilled does not exempt you from it.
The practical version: trust your ears, and then check them with a blind comparison. A genuinely good listener is not someone who is always right. It is someone whose reports survive being tested.
Yes, and pitch discrimination is one of the most trainable listening skills there is. The gains come from a specific kind of practice: short, focused A/B comparisons with immediate feedback on whether you were right. Passive listening — however many hours of it — does almost nothing, which is why lifelong music fans do not automatically test well.
A few things that transfer quickly. Detune a synth oscillator by known amounts and try to name the size. Tune an instrument by ear against a drone and check yourself with a tuner. Practise with narrow-band EQ boosts until you can identify the frequency of a boost by sound alone. And retake a discrimination test like this one every couple of weeks rather than daily, so you are measuring your ear rather than your memory of the last run. The ear training guide covers structured practice across every dimension, not just pitch.
One thing worth knowing before you worry about age: audiometric hearing loss and discrimination skill are close to independent. Plenty of the most trusted listeners in professional audio are well into middle age with measurably reduced high-frequency sensitivity, and they still outperform twenty-year-olds on the judgements that matter. Knowing what to listen for beats hearing more.
An audio industry term for a listener who reliably detects differences most people miss. There is no certification and no fixed threshold — in practice it describes trained professionals whose discrimination has been sharpened by years of deliberate comparison.
With two tones played in sequence, untrained listeners usually become unreliable between 10 and 25 cents. Trained musicians and engineers commonly hold accuracy down to about 5 cents. Below roughly 3 to 4 cents, reliable sequential discrimination is rare outside a lab. Simultaneous tones are a different and much easier task, because beating gives the difference away.
Real for real differences, unreliable for imaginary ones. Blind research consistently shows trained listeners outperforming untrained ones on measurable differences, and equally consistently shows confident claims collapsing when the differences are not there and the test is properly blinded.
No. Detection sensitivity — the quietest sound you can hear — is largely unrelated to discrimination skill. Golden ears are an attentional and perceptual skill, not an audiometric one.
One hundredth of an equal-tempered semitone, so 1200 cents to the octave. It is a ratio unit, which is what makes it useful: 10 cents means the same perceived difference in every register, while 10 Hz does not.
Because simultaneous tones produce audible beating, and counting beats is a different skill from hearing pitch. Playing the tones one after another forces you to hold the first in memory and compare — which is what pitch discrimination actually is, and what makes 4 cents genuinely hard.