Mediatavio
YouTube to MP3PriceBlogPrivacy

Audio analysis

Key BPM Finder: From Audio to Musical Key and Tempo

A practical workflow and the traditional signal-processing pipeline behind the estimates.

By Mediatavio Engineering

What this tool tells you

key bpm finder estimates a track's musical key and major or minor scale, the corresponding Camelot code, and tempo in beats per minute. These estimates are useful starting points for arranging a set or comparing recordings; they are not a substitute for listening.

For example, C major maps to 8B and A minor maps to 8A. Camelot is a lookup derived from the estimated key, not a second independent analysis that confirms it.

Analyze a recording

  1. Open Key BPM Finder and choose audio files, or drop them into the upload area. WAV is useful when a browser cannot decode another format.
  2. Select Analyze audio. Files run sequentially, so one failed file does not stop the remaining queue.
  3. Read the key, Camelot and BPM columns. An uncertain result or a missing metric means the algorithm did not produce an estimate meeting our display criteria.
  4. Use the preview control to listen, remove files you do not need, or cancel an active analysis.
  5. Sign in before Export CSV to download the completed results. The report includes duration and the position and length of the analyzed excerpt.

The queue accepts up to 10 files, at most 100 MB per file and 300 MB in total. Recordings must last at least 8 seconds and no more than 10 minutes. Format decoding depends on the browser.

The implementation

Web Audio decodeMono excerpt at 44.1 kHzEssentia WorkerKey, Camelot and BPM

The page decodes the file with Web Audio, averages its channels into mono and selects at most 120 continuous seconds from the middle. If necessary, an OfflineAudioContext resamples that excerpt to 44,100 Hz. The original file is still decoded first, so excerpting limits analysis work rather than guaranteeing low peak memory.

A Float32Array is transferred to a separate Worker. Essentia.js 0.1.3 runs traditional algorithms: KeyExtractor uses tonal features and key-profile matching, while RhythmExtractor2013 uses its multifeature rhythm method with a 40 to 208 BPM search range. This implementation does not download neural-network weights.

Before analysis, the Worker rejects non-finite samples and near-silent input. We display a key only when its strength exceeds 0.1 and it has a valid Camelot mapping. BPM must be positive and finite, with at least four beat ticks and confidence of at least 0.5. These are algorithm-score thresholds, not calibrated probabilities of correctness.

Reliability and cancellation

A track can change key or tempo, and half-time or double-time interpretations can both be musically meaningful. Percussion-only recordings, sparse notes and strong background noise can produce uncertain results. The middle excerpt may not describe an introduction or outro.

Cancel terminates the analysis Worker. A generation identifier prevents an earlier decode or Worker result from overwriting a later task. Completed files remain available, and unfinished files can be analyzed again. Each Worker releases its Essentia vectors before it ends.

CSV cells escape quotation marks and neutralize leading spreadsheet-formula characters in filenames. Account reservations are completed before a downloadable result is issued; failed exports can release their reservation.

Put the results to use

To compare a different key, use Pitch Shifter. To isolate a representative passage, use Audio Trimmer. For an incompatible recording, try Audio Converter. An Audio Spectrogram can help inspect the sound, but its image alone does not establish the key.

Audio remains in the browser for this analysis. Account and usage requests are separate from the signal-processing pipeline. Essentia is distributed under AGPL-3.0; commercial licensing options and deployment obligations should be reviewed before reusing the engine elsewhere.

Technical references