Audio Blind Test
Two blind listening tests in one. Each round plays a mystery sample: vote whether the music clip was full 16-bit or crushed to 8-bit, or whether a hi-hat landed in sync with the kick drum or milliseconds late. Every tally is judged against chance, so a lucky streak won’t pass for golden ears.
Play the mystery sample and vote — the tally starts with your first answer.
Speakers add milliseconds of delay of their own, and room noise masks the quiet details.
One answer proves nothing. 9 of 10 correct is the first score that beats luck at 95%.
Replay the mystery clip as often as you like — but every vote draws a fresh sample.
Sighted listening is unreliable: when you know which version is meant to be “better”, you hear it as better. A blind test removes that bias — each round draws a sample at random, you commit to one vote, and only the tally decides. The bit-depth test pits a music loop at 16 bits (the CD standard, noise floor near −96 dB) against the same loop truncated to 8 bits, which lifts the noise floor to roughly −48 dB — grit and hiss riding on the quietest notes and decays. The timing test plays a kick drum and a hi-hat and asks whether they struck together, at offsets from an easy 100 ms down to a brutal 1 ms.
Both variants come from the same source samples, so the only difference is the one under test: quantization in the bit-depth game, onset offset in the timing game. Practice with the labeled reference clips for as long as you like, then play the mystery sample. Replays are free, but every vote draws a fresh sample, so there are no second guesses. Use headphones — separate speakers sit at different distances from your ears and add milliseconds of their own. Everything is synthesized locally in your browser; nothing is uploaded.
How many correct answers do you need?
Guessing gets about half the answers right, so a raw score means little on its own. The confidence figure is the probability that your score beats pure coin-flipping — 95% is the usual threshold for a claim like “I can hear this”, and a skeptic will want 99%. Ten trials is the minimum for the math to say anything; more trials protect a genuinely good ear from one unlucky miss.
| Trials | Correct for 95% confidence | Correct for 99% confidence |
|---|---|---|
| 10 | 9 | 10 |
| 15 | 12 | 13 |
| 20 | 15 | 16 |
| 30 | 20 | 22 |
| 50 | 32 | 34 |
What should you expect to hear?
Failing is normal — that is rather the point. The 8-bit version’s quantization noise hides surprisingly well under loud, dense music, which is why 16 bits is already generous as a delivery format, let alone 24. If the bit-depth test feels impossible, nudge the volume up and listen to the tail of the last note, where the music is quietest and the added noise is most exposed. On timing, most people hear a 50–100 ms flam easily, lose the thread between 10 and 20 ms, and almost nobody passes 1–2 ms — work down the ladder and find where your threshold sits.