Web Audio APIwithout the browser

Web Audio graph code can usually be reused. Tone.js integration is tested through the polyfill. Code that depends on HTML media elements or other browser APIs needs adaptation. Load the globals before importing Tone.js:

import 'web-audio-api/polyfill'
const Tone = await import('tone')

const ctx = new AudioContext()
await ctx.resume()
Tone.setContext(ctx)
new Tone.Synth().toDestination()
  .triggerAttackRelease('C4', '8n')

Or start Node with the polyfill already loaded, node --import web-audio-api/polyfill app.js, and a static import * as Tone from 'tone' works.

Also checked through the polyfill: tunajs effects, smplr with its soundfonts fetched over the network, and Pizzicato once its context is resumed. soundfont-player and Howler reach for XMLHttpRequest, navigator, and HTML5 Audio, which Node does not have.

100% of the Web Platform Tests in this project’s Node runner pass: 4,322 of 4,322. This is not W3C certification or a guarantee of complete browser equivalence. See the CI results and test scope and harness adaptations.

Verified for v1.6.0 on 2026-10-12.

The graph API follows Web Audio, including automation and 128-frame rendering. Device access, media elements, and worklet execution differ:

Keep graph construction in one function that takes a context; the browser adapter supplies the user gesture and the native context, the Node adapter imports the package and closes the context when the process is done.

Render in memory, then assert samples, peaks, RMS, spectra, duration, or channel count. No audio device is opened.

import assert from 'node:assert/strict'
import { OfflineAudioContext } from 'web-audio-api'

const ctx = new OfflineAudioContext(1, 44100, 44100)
const osc = ctx.createOscillator()
osc.connect(ctx.destination)
osc.start()

const audio = await ctx.startRendering()
assert(audio.getChannelData(0).some(Boolean))

@audio/assert provides checks for frequency, peak level, clipping, and DC offset. They throw ordinary errors and work with any test runner.

Node has the broadest test coverage. Deno and Bun run offline rendering checks in CI. Other targets have narrower verification, listed below; audio devices and codecs are separate from the DSP engine.

Runtime	Offline render	Speaker	Microphone	Decoding
Node 18+	Yes	Yes	Yes	Yes
Deno 2	Yes	Yes	Untested	Yes
Bun	Yes	Yes	Untested	Yes
Browser tab, Web Worker	Yes	Through the browser’s Web Audio	Untested	Yes
LLRT	Yes	No device	No	Untested
Cloudflare workerd	Yes	No device	No	Untested
QuickJS	Yes, with a shim	No device	No	Untested
JavaScriptCore	Yes, with a shim	No device	No	Untested
Porffor	Compiles and runs	Development builds; conformance and devices unverified
JZ	Compiles and runs	Compiler-revision dependent; conformance and devices unverified

Node runs the full suite and 4,322 WPT in CI. Deno, Bun, and LLRT render offline in CI; Deno and Bun devices and decoding were checked on macOS, the browser row by loading esm.sh in Chromium, workerd and the bare engine shells with the engine bundled and its adapters stubbed. The shim a bare shell needs is EventTarget, Event, queueMicrotask, and timers. Not yet tried: Hermes, GraalJS. See compiler results and reproductions for JZ and Porffor.

The engine also runs as ESM in a browser tab or Web Worker. Workers have no native Web Audio API; use the package’s OfflineAudioContext for rendering there.

import { OfflineAudioContext } from 'https://esm.sh/web-audio-api'

Realtime playback is supported. The simple graphs below render faster than realtime, but offline throughput does not measure playback latency. AudioWorklets run on the main thread, without realtime thread isolation; benchmark your workload if low latency is essential.

Yes. Modules can be registered by URL, data URL, Blob URL, or an inline callback in Node. Worklets currently execute synchronously rather than on an isolated realtime thread.

import { AudioContext, AudioWorkletNode } from 'web-audio-api'

const ctx = new AudioContext()
await ctx.resume()
await ctx.audioWorklet.addModule(scope => {
  class PassThrough extends scope.AudioWorkletProcessor {
    process(inputs, outputs) {
      outputs[0][0].set(inputs[0][0] || [])
      return true
    }
  }
  scope.registerProcessor('pass-through', PassThrough)
})

const node = new AudioWorkletNode(ctx, 'pass-through')

@audio/speaker provides speaker output through a native device backend. The DSP engine itself needs no native bindings. Microphone examples use the optional @audio/mic adapter, then feed PCM into a MediaStreamTrack. You can also set sinkId to any Node writable stream. Speaker output works on Node, Deno, Bun, and through the browser's own Web Audio; the microphone adapter is verified on Node.

decodeAudioData() uses @audio/decode: the format is detected and its codec loaded on demand, without FFmpeg or native bindings, independently of the browser’s built-in codecs.

Compressed

MP3, AAC, Ogg Vorbis, Opus, WMA, AMR, Musepack, QOA

Lossless

FLAC, ALAC, WavPack, APE, TTA

Uncompressed

WAV, AIFF, CAF, DSF and DFF (DSD)

Surround

AC-3, E-AC-3, DTS

Video containers

MP4, M4A, MOV, M4V, 3GP, WebM, MKV, AVI: the audio track is extracted

Tracker modules

MOD, XM, S3M, IT

Codec packages have separate licences, including GPL and LGPL. Check the decoder licence table before redistributing them. Decoding loads the complete audio into memory.

The npm archive (167 KB gzip) contains the engine, TypeScript definitions, and examples. Dependencies are additional, including @audio/decode and its codecs. Codecs are installed together but loaded only when needed; disk usage varies by platform and dependency version.