Voice Chat Help

Display this help text to the user:

claude-talk - Voice conversation with Claude Code (macOS only)

Quick Start

/claude-talk:install    # One-time setup + personalization
/claude-talk:start      # Start talking!

Commands

Requirements

• macOS with Apple Silicon (M1/M2/M3/M4)
• Python 3.12
• Working microphone

Configuration

Settings are in ~/.claude-talk/config.env. Key settings:

• AUDIO_DEVICE - Mic index (find with python3 -c "import sounddevice; print(sounddevice.query_devices())")
• MIC_GAIN - Mic gain multiplier (built-in mic needs ~8.0, USB mics ~1.0)
• KOKORO_VOICE - Kokoro TTS voice ID (e.g., bm_daniel, af_heart)
• KOKORO_SPEED - TTS speed multiplier (default 1.0)

Personalities

Manage multiple personalities with /claude-talk:personality:

• /claude-talk:personality — list all saved personalities
• /claude-talk:personality create — create a new personality
• /claude-talk:personality switch <name> — switch active personality
• /claude-talk:personality edit [name] — edit a personality
• /claude-talk:personality delete <name> — delete a personality
• /claude-talk:personality export [name] — print personality file
• /claude-talk:personality import — import from markdown or file

Personalities are saved in ~/.claude-talk/personalities/. The active one is copied to ~/.claude-talk/personality.md.

Teammate Messaging

Once teammates are spawned, they can send each other text messages:

claude-talk message send <personality> "Hey, how's it going?"
claude-talk message broadcast "Listen up team!"

Messages arrive in the recipient's session as:

• Teammate <name> said: <text> (direct message)
• Teammate <name> said to the team: <text> (broadcast)

Barge-In (interrupt TTS by speaking)

Requires BlackHole 2ch virtual audio device:

1. brew install --cask blackhole-2ch
2. Open Audio MIDI Setup → + → Create Multi-Output Device
3. Check "Built-in Output" (first) and "BlackHole 2ch" (no drift correction)
4. Set Multi-Output Device as system output

BlackHole is auto-detected. When available, you can interrupt TTS mid-sentence by speaking.

Troubleshooting

• No speech detected: Check AUDIO_DEVICE index and MIC_GAIN
• Whisper hallucinations: Increase MIC_GAIN or adjust VAD_AGGRESSIVENESS (0-3)
• Barge-in not working: Ensure system output is set to Multi-Output Device (not just speakers)
• Server won't start: Check if another audio-server process is running