# voice-input-settings.screen

S2 settings category — voice + input config. Implements per [brief S §S2](./S-settings-ia.brief.md). Reached from settings-root.screen.md "Voice & input" card. Voice: working.

## Layout (full-screen sheet)

```
┌─────────────────────────────────────────────────┐
│ ◄ Settings                              Done    │ 56pt
├─────────────────────────────────────────────────┤
│                                                 │
│  ─── Speech-to-text ───                         │
│  Recognizer: ● iOS native ○ Cloud (cocotte.io)   │
│  Auto-punctuate: ☑                              │
│  Auto-stop on silence: [ 2s ▼ ]                 │
│                                                 │
│  ─── Text-to-speech (Cocotte's voice) ───       │
│  TTS engine: [ system-default ▼ ]               │   speech-synthesis MCP picks
│  Voice character: [ warm-alto ▼ ]               │   per available voices
│  Speak action receipts: ○ Always ● High-stakes ○ Off│
│  Speed: ━━●━━ 1.0×                                │
│                                                 │
│  ─── Mic / push-to-talk ───                     │
│  Tap behavior: ● Push-to-talk ○ Tap-and-release  │
│  Long-press behavior: [ Cook-mode ▼ ]            │
│  Voice trigger word ("hey cocotte"): ☐ enabled  │   opt-in
│  ⓘ Background mic listening uses more battery.  │
│                                                 │
│  ─── Composer input ───                         │
│  Default keyboard: [ system ▼ ]                  │
│  Auto-correct: ☑   Predictive text: ☑           │
│  Markdown shortcuts (`**`, `_`, `>`): ☑          │
│                                                 │
│  ─── Voice register override ───                 │
│  ○ Auto (let Cocotte pick) ● Hearth ○ Working ○ Plain│
│  ⓘ Forces all Cocotte copy to one register.    │
│                                                 │
└─────────────────────────────────────────────────┘
```

## Components

| Component | Notes |
|---|---|
| STT | Recognizer choice (iOS-native = on-device; cloud = `cocotte.io` cloud transcription for hands-busy fidelity). |
| TTS | Engine + character + when-to-speak + speed. Hands off to [speech-synthesis MCP](./_engineering-v2-port-map.md). |
| Mic / PTT | Push-to-talk vs tap-and-release default; long-press routes to cook-mode or alternate. Voice-trigger opt-in (battery impact disclosed). |
| Composer input | Keyboard, autocorrect, predictive, markdown shortcuts. |
| Voice register override | Auto / hearth / working / plain. Auto = brief T register-by-surface; override = global. |

## States

1. **Default (iOS-native STT, system TTS)** — privacy-leaning defaults.
2. **Cloud STT selected** — banner: "Audio is sent to cocotte.io for transcription, retained 30s. Override at any time."
3. **Voice trigger on** — banner: "Background mic on. Battery impact: ~5% / day. Disable if unsure."
4. **Voice register forced** — chip on settings-root.screen.md: "Voice override active."
5. **VoiceOver** — same content read; mic settings have additional VoiceOver hints.

## Interactions

- **Tap recognizer radio** → confirm sheet on cloud-STT switch (privacy disclosure).
- **Tap TTS voice picker** → speech-synthesis MCP voice list.
- **Tap "Speak action receipts"** → updates listen-rule for chat-home receipts.
- **Tap voice register override** → commits; chat-home banner reflects.

## Edge cases

- **No microphone permission** — banner: "Cocotte doesn't have mic access. Grant in iOS Settings to enable voice."
- **TTS voice unavailable** (offline) — fallback to system default with chip "voice unavailable, using system."
- **Reduced-haptics + mic active** — single-haptic on mic-start; no haptic on each word recognized.

## Related
- [Brief S §S2](./S-settings-ia.brief.md) — parent.
- [`voice.md`](./00-system-voice.md) — voice register source.
- [Brief J](./_engineering-v2-port-map.md) — speech-synthesis MCP.
- [Brief X](./X-accessibility.brief.md) — accessibility override layer.
- [chat-home.screen.md](./chat-home.screen.md) — composer mic affordance source.

## Out of scope
- Per-language STT (English-only at P0).
- Custom TTS voice training (defer).