Buscar archivos
or drag and drop it anywhere in this area
No file size limit · minimum 10 seconds · 30+ seconds recommended for analysis
Files stay on your device — no uploads · no account · private processing in this browser tab.
Cómo funciona
Every file is analysed locally with a chromagram + onset detection pipeline. Export the table as CSV for your DJ library.
- 01
Upload
Pick a file or drop it in — decoding happens on your device.
- 02
Process
The spectral engine renders the result in this browser tab.
- 03
Download
Preview and save the output as a lossless WAV file.
How to read the results
Every reading carries a confidence rating instead of a bare number. Tempo is measured on several sections of the track and cross-checked against a beat grid, so a half-time or double-time reading is reported as an alternative rather than silently chosen for you. Key detection compares the track's pitch-class profile against three published tonal profiles and reports the relative key as a runner-up, because a major key and its relative minor share the same notes.
The Camelot code (for example 8B) and Open Key code let you find harmonically compatible tracks: neighbours on the wheel, and the relative major/minor sharing the same number, mix smoothly. Tuning is reported in cents against A440 — older records and analogue tape often sit a few cents off, which is why a key can sound "between" two notes.
When the analyser cannot reach a defensible answer — silence, ambient material without a pulse, or clips shorter than a few seconds — it reports Unknown and explains why. That is deliberate: a wrong BPM costs you more time than no BPM.
Spectral engine
Frequency-domain processing tuned for voice and instrument content.
Fast
Rendered locally in this tab — no upload waiting or queue.
Free
Unlimited tracks, no account and no usage limits.
Analysis and separation results are estimates produced by signal processing, not definitive verdicts. Quality depends on the source recording.
How key and tempo detection works here
Tempo is estimated from an onset-strength envelope: the audio is downmixed and resampled, a spectral flux curve highlights note attacks, and autocorrelation over several overlapping windows proposes candidate periods. Candidates are scored by how well a click grid aligns with the strongest onsets, and the fastest grid that fits within an equivalence threshold wins — which is what stops the usual half-time and double-time mistakes.
Key detection builds a chroma profile, estimates the tuning offset in cents so slightly sharp or flat recordings are not misread, and correlates the profile against three published tonal templates (Krumhansl–Schmuckler, Temperley and Shaath). Agreement between templates feeds the confidence score, and the runner-up keys — usually the relative minor or the dominant — are listed so you can sanity-check ambiguous material.
Every result ships with a confidence badge rather than a false sense of certainty. Ambient pieces without a clear pulse, rubato performances and very short clips are reported as low confidence or unknown, which is more useful than a confident wrong number.
How to use the key & bpm finder
- 01Add one or many audio files — analysis runs in a Web Worker so the page stays responsive.
- 02Read the key, Camelot code, Open Key code, BPM and tuning offset for each track.
- 03Check the confidence badge, and expand alternatives when a result is ambiguous.
- 04Export the whole batch to CSV for your DJ library or session notes.
Frequently asked questions
On steady-tempo material the estimate is typically within one BPM, and the grid-alignment stage resolves half- and double-time confusion. Free-tempo or rubato recordings are flagged as low confidence instead of being forced into a number.
Camelot notation maps each of the 24 keys onto a wheel — 8A, 9B and so on — so DJs can find harmonically compatible tracks by moving one step around the wheel or switching letter at the same number.
Not every recording sits at A=440 Hz. Measuring the offset first stops a track tuned 40 cents flat from being reported a semitone away from its real key.
Yes. Drop in multiple files and each is analysed in turn, then export the table as CSV.