feat(voice): Add provider-agnostic TTS support

[Steffen025]

Summary

This PR adds three voice system improvements:

1. Provider-agnostic TTS support - Choose between ElevenLabs and Google Cloud TTS
2. Fix voice spam from parallel agents - Prevent Explore/Plan agents from triggering voice output
3. Runtime volume control - Adjust volume without restarting the server

---

1. Provider-Agnostic TTS Support

Problem

The current voice system assumes ElevenLabs as the only TTS provider. Users who want to use Google Cloud TTS cannot easily switch providers.

Solution

• Add TTS_PROVIDER environment variable detection in getVoiceId() function
• Support GOOGLE_TTS_VOICE_ID for Google Cloud TTS configuration
• Maintain full backward compatibility with ElevenLabs (default behavior unchanged)

Usage

ElevenLabs (default - no changes needed):

# Existing setup continues to work

Google Cloud TTS:

TTS_PROVIDER=google
GOOGLE_TTS_VOICE_ID=en-US-Chirp3-HD-Charon  # Optional, uses server default if empty

---

2. Fix Voice Spam from Parallel Agents

Problem

Generic fallback patterns in extractCompletionMessage() caught ANY completion message, including from Explore/Plan agents. This caused voice spam when running parallel research agents.

Solution

1. Remove generic fallback patterns - voice output is now opt-in only
2. Add silent tier check for native agents (Explore, Plan)
3. Keep agent-specific patterns (🗣️ AgentName: and COMPLETED: [AGENT:type])

Voice Tier System

Tier	Agent Type	Voice Output
Silent	Explore, Plan (native)	❌ No voice
Voiced	Engineer, Architect, etc.	✅ Yes (via [AGENT:type])
Main	Main agent (PAI)	✅ Yes (fallback remains)

---

3. Runtime Volume Control

Problem

Volume settings are read once at server startup and cached. Users cannot adjust volume without restarting the voice server.

Solution

• getVolumeSetting() now reads from ~/.claude/VoiceServer/volume.json on each call
• New CLI tool VoiceServer/Tools/voice-volume.ts for runtime volume adjustments
• Changes take effect immediately (no server restart needed)
• Priority: volume.json > voices.json default_volume > 0.5 (default)

Usage

bun run VoiceServer/Tools/voice-volume.ts           # Show current volume
bun run VoiceServer/Tools/voice-volume.ts 0.3       # Set to 30%
bun run VoiceServer/Tools/voice-volume.ts up        # +10%
bun run VoiceServer/Tools/voice-volume.ts down      # -10%
bun run VoiceServer/Tools/voice-volume.ts mute      # 0%

---

Changes

File	Change
Packs/pai-hook-system/src/hooks/lib/identity.ts	Provider-aware getVoiceId()
Releases/v2.3/.claude/hooks/lib/identity.ts	Same update for release version
.env.example	Document new TTS configuration options
Packs/pai-hook-system/src/hooks/AgentOutputCapture.hook.ts	Remove generic patterns, add silent tier
Releases/v2.3/.claude/hooks/AgentOutputCapture.hook.ts	Same fix for release version
Packs/pai-voice-system/src/VoiceServer/server.ts	Fresh-read volume from config
Releases/v2.3/.claude/VoiceServer/server.ts	Same update for release version
Packs/pai-voice-system/src/VoiceServer/Tools/voice-volume.ts	NEW - Volume control CLI
Releases/v2.3/.claude/VoiceServer/Tools/voice-volume.ts	NEW - Volume control CLI

Test plan

Provider-agnostic TTS

• [ ] Existing ElevenLabs setups work without changes
• [x] Setting TTS_PROVIDER=google switches to Google TTS
• [ ] Voice notifications play correctly with both providers

Voice spam fix

• [x] Explore agents don't trigger voice output
• [x] Plan agents don't trigger voice output
• [x] Engineer/Architect agents still get voice (via [AGENT:type] pattern)
• [x] Main agent fallback still works

Volume control

• [x] voice-volume.ts CLI shows current volume
• [x] Volume changes take effect without server restart
• [x] Priority chain works: volume.json > voices.json > default

🤖 Generated with Claude Code