@polyhymnia/web-audio
v0.4.0
Published
Web Audio playback of MIDI note events: event builders, synth, sampler and player. No dependencies.
Maintainers
Readme
@polyhymnia/web-audio
Web Audio playback of MIDI note events: pure event builders on ., synth and player on ./webaudio, sample player on ./sampler. No dependencies.
pnpm install
pnpm build
pnpm typecheck
pnpm testReleases are versioned with changesets and published to npm as @polyhymnia/web-audio.
MIT licensed.
The ./webaudio entry also provides sample-free rhythm voices: clickInstrument(ctx, { out }) and percussionInstrument(ctx, { out, sound: 'kick' | 'snare' | 'woodblock' }). Percussion uses a 1 ms attack and at most 100/75/45 ms decay respectively, ignores MIDI pitch, supports velocity and stopAll(), and runs without timers. A host can use a separate percussion instrument for input feedback without replacing its player's scheduled metronome or clock.
Microphone audio-session policy
Hosts sharing playback and microphone capture can call setAudioSessionPolicy('play-and-record') from @polyhymnia/web-audio/webaudio. It returns { supported, restore }, feature-detects the browser API, and works before or after the shared context exists. Restore when the host releases capture; restoration is conditional so an older handle cannot overwrite a newer host policy. The host coordinates global policy ownership. This API neither requests microphone access nor captures sound. Existing playback creation keeps its default policy when no explicit host policy is active.
Sample player
createSampler(ctx, { out, samples, fallback? }) from @polyhymnia/web-audio/sampler loads nothing up front. warm(signal?) loads in the background, one sample at a time at low network priority and coarse to fine: one per octave (it resolves then, when every note can play), then one every third semitone. prepare(midis) loads given notes' own samples and canPlay(midis) tells whether they can play now. A note never waits: until its own sample is loaded it borrows the nearest loaded one within a tritone, repitched, or plays on fallback, and a note longer than its sample releases before the audio runs out. loadSampler(ctx, { out, samples }) resolves once every sample is loaded.
