Spec — Sound design vocabulary

*tatus* v0.1.0 Draft (20260522). Owner sign-off required for timbre and final samples; Web Audio synth ships as slice 1.

R1 — The 8-cue catalog

The vocabulary is *losed*— only these 8 cues are canonical. A product needing a cue NOT in this list must amend the spec, not ship a one-off.

*efaults* success, error, voice-wake, voice-end are ON by default; focus, hover, press, notify are OFF. Per R4 the user can flip any of the 8 via Settings.

R2 — Timbre family

The Koder sound vocabulary is a *ingle-octave synth palette*— all cues share the same timbre, only the pitch / envelope changes. Owner curates the timbre; v0.1 advisory default is "soft sine with a short attack and ~50 ms tail." No samples ship until owner picks.

Slice 1 (this spec's first publication) ships *eb Audio API synthesized cues*using deterministic parameters (see § R3 token format). When owner ratifies the timbre, real .wav recordings replace the synth path.

R3 — Token format

Each cue is addressed by a CSS-like custom property and a SDK key:

--kds-sound-<cue>:           url(/sound/<cue>.wav)
--kds-sound-<cue>-duration:  <ms>
--kds-sound-<cue>-gain:      <0..1>

Slice 1 synth equivalents (Web Audio API):

const cues = {
  focus:      { freq: 660, dur: 60,  type: 'sine' },
  hover:      { freq: 520, dur: 60,  type: 'sine' },
  press:      { freq: 440, dur: 80,  type: 'sine' },
  success:    { freq: 880, dur: 180, type: 'sine', glide: [880, 1320] },
  error:      { freq: 220, dur: 180, type: 'triangle', glide: [330, 220] },
  notify:     { freq: 660, dur: 220, type: 'sine', glide: [660, 880] },
  'voice-wake': { freq: 880, dur: 140, type: 'sine', glide: [880, 1320] },
  'voice-end':  { freq: 880, dur: 140, type: 'sine', glide: [1320, 880] },
};

(Numbers above are the *anonical synth slice 1*values. Real .wav ratification swaps the implementation but the public API stays.)

SDK-side:

// koder_kit
await KoderSound.play(KoderSoundCue.success);

// koder_web_kit
import { KoderSound } from '@koder/web-kit';
KoderSound.play('success');

R4 — Mute contract

The user controls sound playback via Settings → "Sounds":

Implementation rules:

1. Master mute MUST be honored — even synthesized cues stop.
2. Per-cue overrides MUST persist across sessions (i18n parity —

   user picks once on any device, applies on every device per the sync model in koder-app/behaviors.kmd § Settings sync).

3. The OS-level mute (silent switch on iOS, "do not disturb" on

   Android) MUST take precedence over Koder's master mute (cannot override the OS).

4. Voice cues (voice-wake, voice-end) MUST also respect the

   voice/wake-word.kmd voice.enabled toggle — disabling voice silences these even when the per-cue setting is ON.

R5 — A11y contract

Sound is *lways*a secondary channel. Every cue listed above MUST have a visual twin (per the persona work in accessibility/personas.kmd § P7 + P8):

Audit fails if a cue plays without its visual twin.

R6 — Implementation surfaces

Cross-platform consumers SHOULD load the canonical Settings tile (KodeVoiceSettingsTile extended with the Sound group) instead of hand-rolling toggles.

R7 — Tests of the contract

R8 — Cross-references

• voice/wake-word.kmd — voice-channel sound handled by Talk Mode

  pipeline; the two voice-* cues here are the boundary markers.

• errors/user-facing-messages.kmd — error cue MUST pair with a

  toast whose copy follows that spec.

• accessibility/personas.kmd § P7 + P8 — Deaf + situational

  hearing personas — visual twin is the load-bearing channel.

• koder-app/behaviors.kmd § Settings — Settings drawer "Sounds"

  group integration.

• policies/security.kmd § media-src — selfhostedonly audio

  files; no third-party CDN.

• services/ai/voice/ — Voice consumer of the voice-wake /

  voice-end cues.

R9 — Open questions

1. Should notify default ON when the OS notification permission

   was granted? (Microsoft Fluent default is ON for systemtoast accompanied sound; KDS default is OFF until user opts in.)

2. Should the SDK expose a play(cue, { gain: 0.3 }) runtime gain

   override, or is the user-level Settings gain the only control?

3. Wave 2 — does sound vocabulary extend to *mbient music*(login

   screen background loop) or stay strictly transactional? Out of scope here; tracked as potential specs/sound/ambient.kmd follow-up.

Sign-off