Written by Hurst | Last updated: June 2026
What This Page Explains
This page documents the exact technical mechanism behind each tempo detection mode on MusicTempoFinder.com — tap tempo, microphone audio analysis, and file upload analysis. It explains what each pipeline does step by step, what accuracy each achieves, and what no tempo detection method can determine. For plain-language answers about what your result means in practice, see the FAQ.
What BPM Actually Measures
Beats per minute (BPM) is the number of beat-level pulses in one minute of music. It is a measurement of tempo — the speed of the underlying rhythmic pulse. BPM is not the same as time signature: a track in 3/4 time and a track in 4/4 time can both be at 120 BPM. BPM does not capture groove, swing, or expressive timing variation. It is a single number that describes the average rate of the beat pulse — a precise and useful measurement with clearly defined scope.
Common tempo ranges by genre:
| Genre | Typical BPM Range |
|---|---|
| Hip-hop / trap | 60–100 BPM |
| R&B / soul | 60–110 BPM |
| House | 120–130 BPM |
| Techno | 130–150 BPM |
| Drum and bass | 160–180 BPM |
| Ambient / downtempo | 60–90 BPM |
| Pop | 100–130 BPM |
| Rock | 100–140 BPM |
These are reference ranges. Genre tempo boundaries are not fixed and individual tracks vary widely.
Mode 1 — Tap Tempo Pipeline
Step 1 — Tap Event Recording
Each tap — whether a keyboard keystroke, a mouse click, or a screen tap — is recorded with a high-resolution timestamp using the browser’s performance.now() API, which provides sub-millisecond precision independent of system clock adjustments.
Step 2 — Interval Calculation
The interval between each consecutive pair of taps is calculated in milliseconds:
Interval (ms) = Timestamp of tap N − Timestamp of tap N−1
Step 3 — Outlier Filtering
Intervals that deviate significantly from the running average — typically more than ±25% — are flagged as likely errors (a misplaced or missed tap) and excluded from the averaging pool. This prevents a single incorrect tap from distorting the result significantly.
Step 4 — BPM Calculation
The BPM result is calculated from the average of the accepted intervals:
BPM = 60,000 ÷ Average interval (ms)
Worked example: If the accepted intervals are 500 ms, 498 ms, 502 ms, and 499 ms:
Average interval = (500 + 498 + 502 + 499) ÷ 4 = 499.75 ms
BPM = 60,000 ÷ 499.75 = 120.06 BPM → displayed as 120 BPM
Step 5 — Session Reset
The session resets automatically after a pause of approximately 2–3 seconds with no new tap input, clearing the interval history and starting fresh. This prevents intervals from a previous tapping session contaminating a new one.
Tap tempo accuracy by tap count:
| Number of Taps | Typical Accuracy |
|---|---|
| 1–3 | Unreliable — insufficient data |
| 4–5 | ±2–3 BPM — usable for approximate reference |
| 6–7 | ±1–2 BPM — reliable for most purposes |
| 8+ | ±1 BPM — maximum reliability under consistent tapping |
Mode 2 — Microphone Audio BPM Detection Pipeline
Step 1 — Microphone Permission and MediaStream Capture
The browser requests microphone permission via getUserMedia(). Once granted, a live MediaStream is opened and connected to a MediaStreamAudioSourceNode within an AudioContext. The audio stream is processed entirely within the browser — it is never recorded, stored, or transmitted. For full details on audio data handling, see the Data Security page.
Step 2 — Energy Envelope Onset Detection
The incoming audio signal is analysed in short overlapping frames using an AnalyserNode. For each frame, the RMS energy is calculated:
RMS = √(Σ(sample²) ÷ N)
A significant increase in RMS energy between adjacent frames — an energy envelope onset — indicates the likely start of a beat. Spectral flux analysis, which measures the rate of change in the frequency spectrum between frames, is used alongside energy onset detection to improve reliability on tracks where beats are marked primarily by harmonic changes (such as hi-hat patterns) rather than low-frequency transients.
Step 3 — Onset Time Series Assembly
The detected onset times are assembled into a time series — a list of timestamps at which beats are estimated to have occurred. This time series is the input to the tempo estimation step.
Step 4 — Autocorrelation for Beat Period Identification
Autocorrelation is applied to the onset time series to identify the most consistent interval between events — the beat period. Autocorrelation computes the similarity of the time series with a delayed version of itself across a range of lag values. The lag at which the highest correlation occurs corresponds to the beat period in milliseconds.
BPM = 60,000 ÷ Beat period (ms)
Worked example: If the autocorrelation peak is found at a lag of 500 ms:
BPM = 60,000 ÷ 500 = 120 BPM
Step 5 — Half-Time / Double-Time Disambiguation
This is the most critical and most frequently misunderstood step in audio BPM detection. The autocorrelation function produces peaks not only at the true beat period but also at integer multiples and divisions of it. A track at 120 BPM will produce autocorrelation peaks at both 500 ms (120 BPM) and 1,000 ms (60 BPM — half time), and potentially at 250 ms (240 BPM — double time).
The tool evaluates multiple tempo candidates using a scoring function that weights candidates against expected genre tempo ranges, the density of onset events per beat, and the regularity of inter-onset intervals. The highest-scoring candidate is returned as the primary result.
However: this disambiguation is heuristic, not definitive. For tracks with ambiguous beat structure — sparse rhythmic content, complex polyrhythms, or tempo at the boundary between common BPM conventions — the algorithm may return the half-time or double-time value. Users are advised to verify the result by ear before using it in professional production or performance contexts.
Step 6 — Result Display
The estimated BPM is displayed alongside the confidence level of the result and, where half-time/double-time ambiguity is detected, the alternative tempo candidate. The Troubleshooting page documents what to do when the displayed result seems to be half or double the expected value.
Mode 3 — File Upload BPM Detection Pipeline
Step 1 — File Selection and Local Reading
The user selects an audio file (MP3, WAV, AIFF, OGG, or FLAC). The file is read locally using the browser’s FileReader API:
FileReader → readAsArrayBuffer(file)
The file is read into an ArrayBuffer in the browser’s memory. It is never uploaded to any server. The file name and metadata are used for display purposes only and are not transmitted.
Step 2 — Local Audio Decoding
The ArrayBuffer is decoded into an AudioBuffer using the Web Audio API:
AudioContext.decodeAudioData(arrayBuffer)
This decoding occurs entirely within the browser. The decoded audio data exists only in the browser’s active memory for the duration of the analysis.
Step 3 — Offline Analysis
The decoded AudioBuffer is analysed using an OfflineAudioContext — a Web Audio API processing environment that runs faster than real time without audio output. The same onset detection and autocorrelation pipeline used in microphone mode is applied to the decoded audio data.
Step 4 — Session Clear
When the analysis completes, the result is displayed. The ArrayBuffer, AudioBuffer, and all intermediate analysis data are released from memory when the tab is closed or reloaded. The uploaded file never persists beyond the active browser session.
Accuracy Summary
| Mode | Typical Accuracy | Primary Limitation |
|---|---|---|
| Tap tempo (8+ taps) | ±1 BPM | Human timing variability |
| Tap tempo (4–5 taps) | ±2–3 BPM | Insufficient averaging |
| Microphone audio | ±2–5 BPM | Half-time/double-time ambiguity, tempo variation |
| File upload audio | ±1–3 BPM | Half-time/double-time ambiguity, sparse rhythmic content |
What This Tool Cannot Determine
Time signature BPM measures the rate of the beat pulse. Whether that pulse is grouped in twos, threes, fours, or a more complex pattern — the time signature — is not determinable from beat period measurement alone.
Expressive timing and tempo variation Live performances and recordings with intentional tempo variation (rubato, ritardando, accelerando) do not have a single BPM. The tool returns an averaged estimate. The further a performance deviates from a consistent pulse, the less meaningful a single BPM figure is.
Groove and swing Rhythmic feel, swing ratio, and microtiming are not captured by BPM. Two tracks at exactly 120 BPM can feel dramatically different depending on the placement of notes relative to the beat grid.
The “correct” half-time or double-time interpretation As documented above, the algorithm cannot definitively resolve whether the true tempo is at the detected value or its half or double. Ear verification is always recommended for professional use.
Related Pages
- FAQ — plain-language answers about result interpretation and common BPM questions
- Troubleshooting — what to do when your result seems wrong or the tool fails to detect tempo
- Data Security — how microphone audio and uploaded files are handled
- About MusicTempoFinder.com — who runs this site and why it was built
Written by Hurst, founder of MusicTempoFinder.com. Last updated: June 2026.