Key Finder

Musical key detection comes down to one question: which of 24 possible keys — 12 major and 12 minor — best matches the tonal content of a track? A neural network solves this by first building a pitch class profile (also called a chromagram): a fingerprint that collapses every note in the audio onto the 12 chromatic pitch classes regardless of octave. The model is trained on tens of thousands of labelled tracks across genres, learning the subtle distributions that distinguish a true tonic from a passing chord. The hardest pairs are parallel and relative keys — C major and A minor share the same seven notes, so detection sometimes returns the relative minor when the actual key is its relative major (or vice versa). That ambiguity is intrinsic to the audio itself, not a flaw in the algorithm. Results display both standard notation (e.g., G minor) and Camelot notation (e.g., 6A) — the DJ-friendly format. See the Camelot Wheel for full compatibility rules.

For cleaner detection, give the model a clean signal. Stem-separated material works best — an isolated harmonic stem strips away drums and reverb tails that muddy the chromagram. Avoid noisy live recordings, heavily clipped masters, or atonal sections. The algorithm needs roughly 30 seconds of sustained tonal content to converge; very short clips or pure-percussion intros can confuse it. If your source is a full mix and detection looks shaky, run it through the Stem Splitter first and feed the instrumental or vocal stem back in.

The DJ workflow is simple: drop each track into Key Finder, capture the Camelot code, and tag your library so you can sort by key. Once tracks are tagged, harmonic mixing reduces to three safe moves on the wheel. Same code (8A → 8A) is a perfect match — melodies fuse without dissonance. One step up or down (8A → 7A or 9A) shifts energy by a fifth and feels like a natural gear change. Letter swap (8A → 8B) keeps the tonic but flips between minor and major, ideal for a mood lift before a drop. Pair this with BPM Finder for tempo or use the combined BPM and Key Finder to prep an entire set in one pass — tempo plus key is the full preparation checklist for a harmonically tight mix.

Producers and remixers use the same workflow in reverse. Detecting the key of a sample tells you whether it will sit on top of your project's tonic or fight it. When matching loops from different sources, knowing both keys lets you decide whether to transpose a loop or rework your chord progression. For pitch-matching across tracks of different keys, the Pitch Shifter lets you transpose by semitones while preserving tempo — useful when you want to align a vocal acapella in F# minor with an instrumental in G minor, or pull a sample into your project's key without destroying its timing.

A convolutional neural network (CNN) trained on 50,000+ tracks analyzes the audio using a Constant-Q Transform (CQT) that captures harmonic content across the full frequency range. The model outputs probabilities for all 24 major and minor keys.

Over 90% accuracy on isolated instruments and clearly tonal music. Accuracy decreases on atonal content, heavily distorted guitar, or tracks with frequent key changes. The confidence indicator shows how certain the detection is.

Yes. Results display both standard notation (e.g., A minor) and Camelot wheel notation (e.g., 8A) for DJ harmonic mixing. Related: use the Camelot Wheel tool to find compatible keys.

The tool reports the dominant key of the entire track. For songs with modulations, the result reflects the most prominent key. For detailed key-change analysis, try the Song Analyzer which provides time-stamped results.

The CNN distinguishes between relative major/minor pairs (e.g., C major vs A minor) which share the same notes but have different tonal centers. This is one of the hardest problems in music information retrieval — the model uses melodic contour patterns, not just pitch class distributions.

No. The CNN model (1.8MB) runs entirely on your device. Audio is processed locally and never leaves your machine.