Why Key Detectors Disagree and Where They Go Wrong

A key detector does not guess at random. It compares the energy distribution across the twelve pitch classes with reference profiles for twenty-four keys and picks the closest one. It goes wrong where two or more answers describe the recording equally well: relative major and minor, modal scales, or a bassline built from only two notes. Below we go through those cases and show how to check the answer by ear.

Other weak spots of automatic detection, including a key change inside the beat and what to do with a beat outside your range, are covered in how to find the key of a beat. This article is about the mechanics that make two detectors disagree.

What the detector actually compares

The algorithm builds a key profile from the recording: how often and for how long each of the twelve scale degrees sounds. That profile is matched against reference profiles for all twelve major and twelve minor keys, following the Krumhansl–Schmuckler method. The key with the closest match wins.

This is note statistics, not chord analysis. Order is not part of the profile. The detector does not know that a phrase ended on the tonic, only that the tonic sounded often. It is confident where one note clearly dominates, and it wavers where two notes compete or where nothing dominates.

How to read the answer

• Note and mode. A letter without a suffix means major, and a lowercase m means minor. An answer of Am means A minor.
• Camelot code. A number and a letter: 8B is C major, 8A is A minor. The same number marks a relative pair, and such keys usually mix without clashing.
• Compatible keys. A list of keys that combine with the result without a clash.
• BPM. The tempo. The detector searches within 60–200 BPM and prefers 70–180.
• Confidence. The tool returns this value together with the answer. When it is low, treat the result as a hypothesis and check it with one of the methods below.

1. Relative keys: the same notes, a different home

C major (8B on the Camelot wheel) and A minor (8A) use exactly the same seven notes: C, D, E, F, G, A and B. They differ only in which note sounds like home. Note content alone cannot separate them. You need the order of the chords and the place where the music comes to rest.

A handy example is the progression Am–F–C–G. In C major it reads vi–IV–I–V; in A minor it reads i–VI–III–VII. Four identical chords, two valid readings. If the loop settles on C and feels like rest, you are in C major. If it keeps returning to Am, you are in A minor. A detector that ignores order sees the same set of notes and picks by a small margin.

For sample picking the mix-up is almost harmless. Relative pairs sit side by side on the Camelot wheel and mix equally well, so a wrong relative answer still suggests samples that fit. For vocals it matters more. The same melody lands on a different home and sounds different, so check the relative option by ear before you record.

2. Modal scales: when two options are not enough

Dorian, mixolydian and the other modes also use seven notes, but they lean on a different note. G mixolydian (G, A, B, C, D, E and F with a natural F) has the same notes as C major. If the music leans on G and the detector picks C major, it is right about the notes and wrong about the home.

D dorian produces the same notes as C major, too. Its minor colour comes from the minor third (D–F) and the natural sixth (B). The detector may choose either C major or D minor, and both are partly wrong: C major misses the home, and D minor lacks the natural sixth. Treat answers like this as provisional, and check the final note of the phrase.

3. A two-note bassline: too many homes

Imagine an 808 bass moving between A and E. That pair fits twelve of the twenty-four keys: C major, D major, A minor, E minor and eight more. The detector cannot choose between them, so it settles on small margins in the energy distribution: whichever note sounds a little louder or longer.

Major or minor is decided by the third. Without a third, the mode simply is not audible. A fifth says nothing about the mode, so a root-and-fifth bass sounds neutral. Even a single A minor triad (A, C and E) fits six of the twenty-four keys, so a short chord loop does not settle the question either.

The practical rule: if the recording has only two notes, add a third element to your check, such as a chord or a phrase with a third in it. Then the answer has an anchor, and you can compare it with what you hear.

What to do when two detectors disagree

Two detectors can disagree without either one being broken. A recording often allows more than one reasonable answer, and each tool picks one of them, so the useful question is not "which key is right?" but "which options fit, and why was this one chosen?"

Take a two-note 808 bass that moves between A and E. The answer could be A minor (8A on the Camelot wheel) or C major (8B). Both give the same notes, so arguing about the notes gets you nowhere. What settles it is the note the beat comes to rest on.

Start by making sure you are comparing the same thing. Load the same file, unchanged, without re-encoding or trimming. A re-encode can shift the profile a little, and an excerpt of a different length describes a different part of the recording, so answers for two versions of a file cannot be compared.

Then look at which part of the answer differs. If two Camelot codes share a number and differ in the letter, such as 8A and 8B, you are looking at a relative pair: the notes are the same, only the home differs, and the last chord of the phrase decides it. If the numbers differ, the note sets differ as well, so settle the question with the audio itself, using the check described below.

If both answers come with a low confidence value, do not pick either one by the number alone. Take a longer passage, such as a full verse or chorus, and run the detection again. An answer that stays the same across passages deserves more trust; one that changes with every cut does not.

Common mistakes when comparing answers:

• Misreading the letter. Am is A minor, while a bare A is A major.
• Calling a relative pair an error. A minor and C major share every note; only the home is in dispute.
• Checking the answer against a label in the filename or the sample pack. The label may describe another version of the file.
• Treating a low-confidence answer as a fact.

How to check by ear

Two cues usually point to the tonic.

The first is the bass on the strong beat. If the bass repeats one note on the first beat of the bar, that note is usually the chord root. Listen to whether it feels like rest rather than a passing sound.

The second is the end of the phrase. A phrase usually settles on the tonic or on the dominant, which pulls back to the tonic. Loop the last two bars and listen to where the music stops.

If both cues point to the same note, the detector's answer is probably right. If they disagree, you are looking at a relative pair or a modal scale, and the decision should rest on the feeling of home rather than on statistics. The triad test we describe in how to find the key of a song is still worth keeping in your toolkit.

How to do it in Magic Master

• The key and BPM finder is free and needs no sign-in. It returns the note, the mode, the Camelot code, compatible keys, the BPM and a confidence value. The file is deleted after analysis.
• If the answer is right but the beat does not sit under your voice, the key changer shifts a song or a backing track by −6 to +6 semitones and keeps the tempo. Output is MP3 320 or WAV. Free use: 10 shifts a day for guests and 30 for signed-in accounts.
• Before mastering, upload the finished mix to the app. Base mastering runs within quotas: guests get 3 runs per calendar month, and accounts without tokens get 1 per day. Mastering with PRO modules costs one token.