Other tools show you one key and let you find out the hard way if it's wrong. KEYD lights up the wheel — the detected key and its harmonic neighbourhood — so even before it's certain, you can see where the track lives and what mixes with it.
As confidence climbs, keys are filtered down. A single segment shows certainty.
If KEYD gets it wrong, tap any segment to set the key manually.
The wrong key is worse than no key.
KEYD tells you when it's sure — and when it isn't.
Key and tempo are tracked for confidence. Watch the dots go orange → yellow → green as confidence increases. They move in step with the wheel: orange shows three or four candidates, green a single one.
When it's green, trust it. When it isn't, you know to use your ears.
KEYD locks onto the kick, not the hi-hats — so shuffled house, dotted basslines, and busy percussion can't fake it out. Once locked, breakdowns and fills won't shake the number.
If the tempo still doesn't look right, just tap the numbers to input manually.
KEYD uses built-in track recognition to name what's spinning. Wrong match? Obscure white label? Enter your own info and save it.
Every track you save builds a library you can sort and filter by BPM and key. One tap filters to everything compatible with what's playing right now — your next record, from your own crate.
Generic beat detectors treat every frequency equally — so busy hi-hats and syncopated percussion can outvote the kick, producing the classic failures: half-time, double-time, and the "dotted" error where a 120 BPM record reads as 160.
KEYD weights its onset detection toward the kick band, roughly 40–180 Hz — where a four-on-the-floor kick lives — so the tempo is driven by the beat rather than the hats on top of it. It then verifies that the music actually repeats at the winning tempo before trusting it, and explicitly rejects dotted-rhythm aliases. Once locked, breakdowns, fills, and vocals can't shake the reading — and every change to the engine is graded against a corpus of real vinyl recordings before it ships.
Key detection is a matter of judgement, not a lookup — so KEYD doesn't rely on a single method. It runs two independent key-analysis models in parallel: one built on classical tonal theory (the Krumhansl–Kessler profiles from decades of music-cognition research), and one tuned specifically on modern electronic and dance material. Acoustic records and electronic tracks each play to a different model's strengths.
The two run side by side and their results are overlaid. When both models independently agree on the same key, that's a strong signal and confidence rises; when they disagree, KEYD holds back and keeps listening until the evidence resolves. Two theories concurring is far more trustworthy than one guessing.
Field testing revealed something useful: when key detection misses, it almost never misses randomly. The true key is nearly always adjacent on the harmonic wheel — the relative key or a neighbour — which are exactly the keys that mix harmonically anyway.
So instead of one fallible answer, KEYD lights the wheel as a heat map of the harmonic neighbourhood. You get a usable, mixable answer earlier than any single-key display could honestly give you.
It's honest, by design: a blend of how dominant the reading is in the accumulated evidence, how tightly recent readings agree, and whether the two key models concur. The display settles, then only rises as confidence increases. Green means the reading has proven itself; anything less is a signal to analyse again.
KEYD reads what's actually playing, pitch riding included — the BPM you see is the BPM in the room, not a database value. While locked, a live drift hint shows how far the current groove has moved from the locked reading.
No connection required. Key and tempo analysis run entirely on the phone and work offline. Only track naming (artist, title, artwork) uses the network.