How BPM Finder analyzes your audio
Learn how BPM Finder estimates BPM, musical key and Camelot in your browser, what Pro scores mean, and how to check the results with downloadable audio examples.
- Select your audioA file from your device, rather than a song link.
- Analyze in your browserTempo and pitch patterns are measured locally.
- Listen and compareReview BPM, musical key and the derived Camelot code.
Your audio stays on your device.
BPM Finder decodes and analyzes the audio you select inside your browser. It estimates BPM and musical key from the complete decoded recording. Audio files are not uploaded for analysis. The tool does not search a song-title database or analyze Spotify or YouTube links.
Unsaved results stay in the current page session. Refreshing or leaving clears them. You can export CSV for free. If you have Pro and select Save Results before analysis, new result snapshots are saved to your account; the audio is not included. Account services and site analytics are separate from audio processing. See Privacy for those data practices.
What each measurement means
BPM and key analysis use signal measurements. Pro’s Dance and Happy scores use an audio model to compare the recording with text descriptions. These outputs answer different questions.
| Result | How it is estimated | What to check |
|---|---|---|
| BPM · Free | Repeated changes in the audio signal suggest the pace of the underlying beat. | Half-time, double-time and changing tempo can produce a different useful count. |
| Key · Free | Pitch-class energy is compared with major and minor key profiles across the recording. | A whole-track label cannot map every key change. Resonant drums can return a candidate without establishing a musical key. |
| Camelot · Free | The selected musical key is mapped to a Camelot code. | This is a notation conversion, not a second independent measurement. |
| Energy · Pro | Average signal level is mapped to a 0–100 scale. | Loudness in the file is not the same as musical intensity or perceived excitement. |
| Dance / Happy · Pro | Audio/text similarity scores are combined across ten-second windows covering the recording. | They are relative model estimates, not objective mood measurements or calibrated probabilities. |
Free BPM and key analysis does not need the Pro mood model. On first Pro use, your browser downloads model and runtime files; processing time depends on the device and recording length. If the mood model cannot load, Dance and Happy remain unavailable rather than being guessed from tempo or key.
Check the estimate before you use it.
- Count the pulse. Tap along in Tap Tempo or compare the track with the metronome. A result near 70 BPM may describe a beat you prefer to count at 140 BPM.
- Listen to a pitched passage. Compare the proposed tonic and major/minor chord with the melody, bass and phrase endings. Check the section where you intend to mix, especially if the song changes key.
- Compare key names before codes. Relative major/minor keys share notes; enharmonic names can describe the same sounding pitches. The Camelot chart converts a key after you have checked it.
- Edit the label if needed. A track’s menu offers half/double BPM and result editing. These controls update the result; they do not stretch, transpose or rewrite your audio. Export CSV before you leave if you want to keep the current results.
For a detailed musical check, follow four ways to find a song’s key. If a recording contains no clear melody or harmony, a key label alone is not evidence that it has a tonal center.
Try the same files we checked.
These original teaching recordings each last 16 seconds at 120 BPM. One chord lasts two seconds, and the four-chord sequence repeats twice. In the recorded check on , the unedited analyzer returned the following results.
A minor: Am – Dm – E – Am
Observed result: A minor · 8A · 120 BPM. Download A minor WAV.
C major: C – F – G – C
Observed result: C major · 8B · 120 BPM. Download C major WAV.
Download a file, select it in BPM Finder and compare the result before making an edit. The verification record includes the file hashes, recording format, analysis version and reproduction steps. The worked example explains what to listen for.
These two clear, synthesized examples are teaching material, not an accuracy benchmark for real-world music. They do not establish a success rate across genres, live recordings, ambiguous loops or percussion. A different engine version may return different estimates.
Where a single label has limits
- Variable tempo, sparse beats and syncopation can make more than one BPM count plausible.
- Relative keys, key changes, tuning differences and weak tonal evidence can affect key estimates. A drum resonance can favor a pitch profile without implying a composed key.
- Dance and Happy depend on the model’s audio/text matches and the selected descriptions. They are not substitutes for listening or judgments about how a person will feel.
- Each file can be up to 50 MB and 10 minutes long, with up to 20 files per batch. Browser codec support and available memory affect which files can be decoded.
Keep the tab open until analysis finishes. If a file will not decode, try a WAV or MP3 copy you have permission to use. For a technical problem, email support@bpmfder.com with your browser and reproduction steps.
Methods, versions and component sources
The default analysis engine is commercial-clap-2. Original signal-processing code estimates tempo from onset periodicity, key from chroma and key-profile matching, and Energy from RMS signal level. CLAP audio embeddings are used for the optional Pro Dance and Happy comparisons; they do not supply the free BPM or key estimates.
The analysis notes and component credits identify the model, runtime, scoring approach and licenses. The model manifest records component versions and hashes. These technical records explain the implementation; they are not a claim of certification or an independent accuracy evaluation.
The guides and teaching material are published under BPM Finder editorial. Send a correction or a question to support@bpmfder.com.