Audio

The runtime has the Audio element (HTMLAudioElement) and Web Audio. Making sound needs the audio permission in the manifest: without it AudioContext stays suspended, resume() and audio.play() fail with NotAllowedError, and videos are muted. OfflineAudioContext and decodeAudioData need no permission.

// Play a sound from the package
const chime = new Audio('sounds/chime.mp3');
chime.volume = 0.6;
chime.play();

// Web Audio: decode once, loop it, and visualize it with an AnalyserNode
const context = new AudioContext();
const analyser = context.createAnalyser();
analyser.connect(context.destination);
const buffer = await context.decodeAudioData(await (await fetch('sounds/loop.m4a')).arrayBuffer());
const source = new AudioBufferSourceNode(context, { buffer, loop: true });
source.connect(analyser);
source.start();

const levels = new Uint8Array(analyser.frequencyBinCount);
requestAnimationFrame(function draw() {
  analyser.getByteFrequencyData(levels);
  // ...
  requestAnimationFrame(draw);
});

Nodes

AudioBufferSourceNode, OscillatorNode (with PeriodicWave), ConstantSourceNode, GainNode, BiquadFilterNode, IIRFilterNode, DelayNode (in feedback loops too), StereoPannerNode, PannerNode and context.listener, AnalyserNode, DynamicsCompressorNode, ConvolverNode, WaveShaperNode, ChannelSplitterNode, ChannelMergerNode. AudioParam has all the automation methods, and nodes can be connected to parameters for modulation.

There is no AudioWorklet, ScriptProcessorNode, MediaElementAudioSourceNode or media streams. PannerNode's HRTF pans with equal power, like equalpower; WaveShaperNode takes oversample but doesn't oversample.

Notes

• Most audio libraries written for browsers read globals such as window and document, which the runtime doesn't have (there is no DOM). three.js audio works as it is in the template's src/three/. howler.js needs the project to provide stand-ins for window, document, navigator and location; its Web Audio mode loads sounds with XMLHttpRequest, which the runtime has.
• An allowed AudioContext is running right away: there is no user gesture to wait for on the desktop. When the app pauses content (see Pausing) it becomes interrupted and its time stops, then goes back to running; playing Audio elements pause too and continue afterwards. When drawing only stops, or the content is muted, sound isn't affected.
• The src of Audio can be a file in the package (with the same path rules as require) or a defile:// address, and an http(s) address with the network permission; data: addresses aren't supported. It plays and decodes what macOS does: MP3, AAC / M4A, WAV, AIFF, CAF, FLAC, ALAC and more, and Ogg on newer systems; check with canPlayType().
• Sound plays on the system's current output device and follows it when it changes; the default sample rate of AudioContext is the device's.
• Users can set the volume of each item of content in the app, or mute all content. It multiplies the volume the content sets, and the volume JavaScript reads doesn't change.
• When content is removed or stopped, all its sound stops at once.
• Reading the system's audio output (for music visualizers, say) isn't possible yet.