BPM Detection in Node.js

realtime-bpm-analyzer/node runs the offline analysis outside the browser, in Node.js, Bun or Deno. It needs no Web Audio API and adds no dependency: you pass decoded samples and get tempo candidates back.

import {analyzeChannelData} from 'realtime-bpm-analyzer/node';

// channelData: a Float32Array of samples in [-1, 1], from any decoder
const tempos = await analyzeChannelData(channelData, 44100);

console.log('Detected BPM:', tempos[0]?.tempo);

CommonJS works too: const {analyzeChannelData} = require('realtime-bpm-analyzer/node').

In TypeScript, the subpath needs a moduleResolution that reads package exports (node16, nodenext or bundler), and the shared type declarations mention URL, so keep @types/node (or the DOM lib) in the project, as Node projects usually do.

Same results as the browser

analyzeChannelData is the Node counterpart of analyzeFullBuffer. The browser version lowpasses the audio with a Web Audio BiquadFilterNode. The Node version uses a plain JavaScript biquad with the coefficients the Web Audio specification defines for that node, then runs the same peak detection and tempo grouping.

The test suite checks this on every change:

• the filter output matches Chromium's OfflineAudioContext render within 1e-5 per sample (measured at most 1.4e-6), and its magnitude response matches getFrequencyResponse
• on the test fixture and on click tracks from 90 to 174 BPM, both functions return identical candidates: same tempos, counts and confidences

analyzeFullBuffer analyses the first channel of the buffer. Pass channel 0 to reproduce it exactly, or a mono downmix of your own.

Decoding the audio

The entry point takes samples, not files, so you choose the decoder. A WAV reader, an ffmpeg pipe (-f f32le -ac 1) or any package that returns a Float32Array works. Keep the file's sample rate, or pass the rate you resampled to.

Options

The third argument takes the same options as analyzeFullBuffer:

Option	Default	Meaning
frequencyValue	200	Lowpass cutoff in Hz
qualityValue	1	Lowpass Q, in dB as Web Audio reads it
tempoRange	{90, 180}	Range the tempos fold into; max must be at least 2 × min

const tempos = await analyzeChannelData(channelData, sampleRate, {
  tempoRange: {min: 70, max: 140}, // report 87 rather than 174
});

It resolves to up to 5 candidates sorted by count, each with tempo, count and confidence (the Tempo type in the API reference). Silence or audio without a clear beat resolves to an empty array.

It throws when channelData is not a non-empty Float32Array, and throws a RangeError for a sample rate that is not positive, a filter option that is not a finite number, or an invalid tempoRange.

The filter on its own

applyLowpass(channelData, sampleRate, options?) is exported too. It returns a new Float32Array filtered like the browser pipeline, and leaves the input untouched.