BPM & Key Finder

BPM detection uses a neural network ensemble with 1 BPM accuracy. Key detection uses a custom convolutional neural network trained on 50,000+ tracks. Both run locally on your device — no server inference.

DJs and producers almost always need both — knowing the tempo without the key (or vice versa) means loading the file into a second tool. This page runs both analyses in parallel on a single upload, so a set of 50 tracks is prepped in one pass instead of two.

No. Both neural networks run entirely on your device. Your files never leave your machine — not even temporarily for analysis.

Both detectors share one decoded audio buffer — Brizm runs tempo and key analysis in parallel using two neural networks. The combined pass is roughly the same speed as either tool alone. You get BPM, key (major or minor), and Camelot code in a single result.

BPM-matching alone does not guarantee tracks sound musical together. Two tracks at 128 BPM in incompatible keys clash during long blends. Combining BPM and Camelot key lets you pre-filter by harmonic compatibility before building a set. See the Camelot Wheel for compatibility rules.

Same neural networks under the hood — this page runs both in parallel on a single upload, while BPM Finder and Key Finder each focus on one signal at a time. Use the single-purpose tools when you only need one value, or when you want to compare results against another detector. Use this combined tool for DJ set prep, where every track needs both values.

MP3, WAV, FLAC, OGG, M4A and most other common audio formats up to 30MB per file — enough headroom for full-length DJ-friendly mixes and album tracks. Larger files are decoded in chunks rather than rejected. For very long mixes, consider analysing individual tracks instead of the full session.

Beat-matching is necessary but not sufficient

Mixing two tracks at the same BPM keeps the kick patterns aligned, but identical tempo says nothing about whether the tracks belong in the same musical universe. If one track is in F minor and the next is in B major, the harmonic distance between them — six steps on the Camelot wheel — is the maximum possible. Sustained synths, basslines, and vocal phrases collide head-on during the blend, producing the dissonant churn that turns a smooth transition into a car-crash. Skilled DJs work around this by cutting fast or running effects, but those are repairs for an avoidable problem. The fix is to know both values up front: the BPM tells you the tracks can mix, the Camelot code tells you they should.

A workflow built around the combined readout

Drop an entire crate into the queue and let the combined detector tag every file. Sort the library by Camelot code first, then by BPM within each code group, and the natural set order falls out almost by itself: nearby Camelot codes (8A → 8B → 9A) blend cleanly, and adjacent BPM ranges within each group avoid the audible pitch shifts that wide tempo jumps produce. Recherche de BPM and Key Finder give you the same neural networks running individually if you only need one value; this page exists because the DJ workflow almost never calls for one value at a time.

One decode, two neural networks, parallel inference

The expensive part of analysing an audio file is not the detection — it's getting the file into memory as a usable signal. Reading the container, decoding the lossy codec, resampling to the model rate, and converting to a spectrogram representation eats most of the wall-clock time. Brizm decodes once and feeds the resulting buffer to two parallel neural networks: a tempo-tracking model that estimates beat positions and the corresponding BPM, and a key-detection model that builds a pitch class profile and classifies it into one of 24 keys (12 major plus 12 minor). Sharing the decoded buffer means the combined pass is barely slower than running either model alone, even though it produces twice the output.

Camelot notation in plain language

The detector emits both a standard key name ("F# minor", "C major") and a Camelot code (the format DJs use: 11A, 8B, etc). Numbers 1–12 wrap a clock; the letter A indicates minor, B indicates major. Moves that work harmonically: same code, adjacent code (clockwise or counter-clockwise by one), and same number with letter flip (8A → 8B). Moves that sound rough: two or more steps away on the clock, or a flip combined with a step. The Camelot Wheel reference visualises every relationship at once. Once a library is tagged, building a set is a matter of moving short distances on the wheel between tracks of similar tempo.

Beyond DJ prep, the same readout is useful for producers picking samples and remixers matching acapellas. The Pitch Shifter can move a sample by semitones while preserving its tempo, so once you know a sample's key and a project's key, transposing the gap is a single slider.