Two songs at the same tempo can still clash badly. What decides it is the key — and the relationship between two keys is something you can read off a wheel.
Key detection works by counting. The audio is cut into short windows, each window is turned into a spectrum, and the energy in that spectrum is folded down into twelve buckets — one per note name, ignoring which octave it came from. That twelve-number summary is a chroma vector, or pitch-class profile. Add them all up across the track and you have a picture of which notes it actually spends its time on.
Then the profile is compared against a template for each of the twenty-four keys — twelve major, twelve minor — and the closest match wins. The templates come from studies of what actual music does: in C major, C and G dominate, F and A appear often, and F♯ almost never.
If a reading looks wrong, check its relative first: the relative minor of a major key is three semitones down (C major → A minor), and the relative major of a minor key is three semitones up.
Tempo detection looks for regular spikes in how fast the energy is rising — the onsets — and finds the interval that best explains them. It is more reliable than key detection, with one recurring failure: octave errors. A track at 140 BPM with a strong backbeat can read as 70, and a 70 BPM track with busy hi-hats can read as 140. Both readings describe the same music; they just disagree about which pulse is the beat.
The fix is to tap along. If the number you tap is half or double what the detector said, the detector was not wrong so much as reading a different layer, and you can use whichever fits the arrangement you are building.
Arrange the twelve keys so each is a fifth from the last, and neighbours turn out to share six of their seven notes. That is the whole reason the circle is useful: distance on the wheel is dissonance.
DJs often use Camelot notation — 8A, 8B and so on — which is the same circle with numbers instead of key names, so "one step around" becomes "add one" and you do not have to know any theory to use it.
Three ways out, in order of how well they usually work:
Ducker's Mix page loads two tracks, reads the key and tempo of each, places them both on the Circle of Fifths and tells you plainly how far apart they are, what they share, and what to do about it — including which way to transpose and by how much to land somewhere compatible. It will also build a bridge undertone from the notes both keys contain.
The analysis runs in your browser on files you never upload.
Open Ducker and check two tracks → Free, no account. Load two files and read the answer off the wheel.Load it into Ducker’s Mix page: it analyses the file in your browser and reports the key and tempo without uploading anything. If the answer looks wrong, check the relative major or minor, which shares all the same notes.
Almost always because one chose the relative major and the other the relative minor. They contain identical notes, so the choice comes down to which one feels like home, and different algorithms weigh that differently.
How many notes they have in common. Neighbouring keys share six of seven and blend smoothly; keys three or more steps apart share very little and will clash. Relative major and minor share all seven.
Yes, if they are close on the circle. If they are not, transpose one of them toward the other, bridge the transition with a note both keys contain, or make the cut somewhere neither track has pitched material.
Because both readings describe the same music at a different layer. Tap along; whichever pulse you naturally tap is the one to use for what you are building.