Bit-exact BS.1770-4 Integrated LUFS, LRA, and True Peak measurement for the browser, Web Workers, and Node — pulled from luvlang.studio's production mastering chain.

• ✅ BS.1770-4 K-weighted Integrated LUFS with absolute (-70) + relative (-10) gating
• ✅ BS.1771 / EBU R128 LRA — 3 s short-term blocks, 10th–95th percentile
• ✅ BS.1770-5 True Peak — 4× polyphase sinc oversampling (24-tap Blackman-Harris)
• ✅ Mono-sum LUFS for mono-compatibility checks
• ✅ Sample-rate aware — works at 44.1, 48, 88.2, 96, 176.4, 192 kHz (and anything in between)
• ✅ Zero dependencies, pure ES module, ~10 KB minified
• ✅ Browser, Web Worker, and Node — same API everywhere
• ✅ Algorithm validated against a 21-case libebur128 reference harness in production

Live demo: luvlang.studio/free-lufs-check — drop a track in your browser and see the numbers.

Most LUFS libraries on npm are either Node-only (FFmpeg bindings), WASM-heavy (libebur128 ports), or approximations that drift on non-48k content. lufs-web is the same pure-JS implementation that powers a live mastering platform — small enough to ship in a Web Worker, accurate enough to use in production, and free of native dependencies.

npm install lufs-web

import { measure } from 'lufs-web';

// AudioContext.decodeAudioData → AudioBuffer
const audioBuffer = await audioCtx.decodeAudioData(arrayBuffer);

const result = measure({
    sampleRate: audioBuffer.sampleRate,
    channels: [
        audioBuffer.getChannelData(0),
        audioBuffer.getChannelData(1),
    ],
});

console.log(result);
// {
//   integratedLUFS: -14.2,
//   lra: 7.3,
//   truePeakDB: -1.4,
//   truePeakLin: 0.853,
//   monoLUFS: -17.2,
//   monoDelta: 3.0,
// }

import { measureIntegratedLUFS, measureLRA, measureTruePeak } from 'lufs-web';

const lufs = measureIntegratedLUFS({ sampleRate: 48000, channels: [L, R] });
const lra  = measureLRA({ sampleRate: 48000, channels: [L, R] });
const { truePeakDB } = measureTruePeak({ channels: [L, R] });

// worker.js
import { measure } from 'lufs-web';

self.onmessage = ({ data: { sampleRate, channels } }) => {
    const result = measure({ sampleRate, channels });
    self.postMessage(result);
};

// main.js — transfer channel buffers zero-copy
const worker = new Worker(new URL('./worker.js', import.meta.url), { type: 'module' });
worker.postMessage(
    { sampleRate, channels: [L, R] },
    [L.buffer, R.buffer],
);

For reference — what each platform normalizes to:

If your master is louder than the target, the platform applies negative gain at playback — it does not put dynamics back into the audio. A master at −7 LUFS gets ~7 dB of gain reduction on Spotify, ending up quieter and flatter at the listener than a master correctly targeted at −14.

Full BS.1770-4 measurement. Returns:

{
    integratedLUFS: number;  // gated mean per BS.1770-4 §3. -70 for silence.
    lra: number;             // BS.1771 loudness range in LU. 0 for tracks < 3 s.
    truePeakDB: number;      // capped at 0 dBTP. Above 0 = inter-sample clipping.
    truePeakLin: number;     // linear amplitude at the peak
    monoLUFS: number;        // integrated LUFS after L+R mono-sum
    monoDelta: number;       // stereo - mono. Positive = mono is quieter.
}

Just the integrated LUFS. Cheapest path if you don't need LRA or true-peak.

Just the loudness range (LU).

Just the inter-sample peak. No sample rate needed — it's a per-sample calculation.

Low-level building blocks — re-exported in case you want to roll your own gating strategy.

• The K-weighting biquad coefficients are exact to libebur128's reference implementation, derived per BS.1770-4 §2.1.
• Integrated LUFS gating: absolute (-70 LUFS) → mean → relative (-10 LU from mean). Identical to libebur128.
• LRA gating: absolute (-70) → mean → relative (-20) → 10th/95th percentile of the gated short-term distribution. Per EBU Tech 3342.
• True peak: 4× oversampling with a 24-tap sinc kernel windowed by a 4-term Blackman-Harris (-92 dB sidelobes). DC-normalized to unity gain.

The production version this was extracted from is validated against:

• 21 reference signals (BS.1770-4 Annex 1 + EBU Tech 3341 vectors)
• 4 sample rates (44.1, 48, 96, 192 kHz)
• K-weighting cancellation tests (white noise in → flat dB out after k-weight removal)

You can run the included unit tests with npm test.

A 3-minute stereo 48 kHz track measures in ~150 ms on an M1 Mac. The K-weighting filters dominate (linear in samples). For long tracks or low-end devices, run inside a Web Worker.

MIT — see LICENSE. Pulled from LuvLang Studio's production mastering chain with permission.

We use this every day in luvlang.studio to measure every track that runs through our 24-stage mastering chain. The code is small, well-tested, and useful to anyone building audio tools for the browser — so we open-sourced it. PRs welcome.

If you want to hear what a full mastering chain does to your track, drop one at luvlang.studio/app — A/B preview is free.