Speech Automation Plugin

Intelligent speech notifications for Claude Code using Kokoro TTS.

Features

🔊 Manual Speech Commands - /speak, /voices, /speech-status, /speech-config
🤖 Automatic Speech Hooks - Smart notifications for errors, tests, builds, and long operations
⚡ GPU-Accelerated - Fast synthesis with Kokoro TTS 82M model
💾 Smart Caching - 10,000x speedup for repeated phrases
🎭 8 Voices - American and British, male and female
🎯 Priority Scoring - Intelligent event classification (0-100 scale)
⏰ Quiet Hours - Reduce or disable notifications at night (10 PM - 7 AM)

Prerequisites

This plugin requires the speech-tts MCP server to be configured.

The speech-tts MCP server must be installed and configured via:

claude mcp add speech-tts /path/to/Speech_MCP/start-mcp-server.sh

ML server must be running at http://ml-server:8001 with Kokoro TTS

Installation

Install the plugin:

claude plugin install ~/.claude/plugins/speech-automation

Or for local development, the plugin can be loaded from:

~/.claude/plugins/speech-automation/

Commands

/speak

Synthesize speech from text:

/speak "Hello world"
/speak "Build complete" --voice am_adam --speed 1.2

/voices

List all available voices:

/voices
/voices --play-samples

/speech-status

Check system status:

/speech-status

/speech-config

Configure automatic speech settings:

# View current configuration
/speech-config

# Enable/disable automation
/speech-config enable
/speech-config disable

# Adjust settings
/speech-config set min_priority 50
/speech-config set events.errors.enabled false
/speech-config set quiet_hours.start_hour 23

Automatic Speech

⚠️ Hooks are now installed! See HOOKS-SETUP.md for details.

The plugin includes intelligent hooks that automatically trigger speech for important events:

Event Types:

🔴 Errors - Exceptions, failures, crashes (voice: Adam)
⏱️ Long Operations - Commands taking 30+ seconds (voice: Bella)
✅ Test Results - Pytest, Jest, test pass/fail (voice: Emma)
🏗️ Build Completions - Webpack, Vite, builds (voice: Sarah)
🔀 Git Operations - Commits, pushes, merges (voice: George)

Getting Started:

# Enable automatic speech
/speech-config enable

# Test with a long operation
sleep 35 && echo "Done"

# Adjust sensitivity
/speech-config set min_priority 40  # Higher = fewer notifications

Configuration

ML Server: http://ml-server:8001 Text Limit: 800 characters (optimal quality) Cache: 100 entries, 1 hour TTL Audio Location: ~/.cache/speech-mcp/

Voices

af_bella - American Female (warm, friendly)
af_sarah - American Female (professional)
am_adam - American Male (calm, measured)
am_michael - American Male (energetic)
bf_emma - British Female (refined)
bf_isabella - British Female (elegant)
bm_george - British Male (distinguished)
bm_lewis - British Male (modern)

Performance

Cache Hit: ~0.01ms
Synthesis: 200-500ms (GPU)
Audio Playback: ~4ms

Requirements

ML server running on http://ml-server:8001
Kokoro TTS service deployed
Claude Code >= 1.0.0

Version

1.0.0 - Full release with manual commands and automatic speech hooks

Implemented:

✅ Manual speech commands (/speak, /voices, /speech-status, /speech-config)
✅ Automatic speech hooks (PostToolUse, SessionStart, Stop)
✅ Quiet hours configuration (10 PM - 7 AM)
✅ Priority-based notifications (0-100 scoring)
✅ Event-driven speech automation (errors, tests, builds, git, long-running)
✅ Smart caching (10,000x speedup)
✅ GPU acceleration (RTX 5070 Ti)
✅ 8 voices (American and British, male and female)

Future Enhancements:

Voice emotion control (pitch, tone variation)
Custom voice cloning
SSML support for rich speech formatting
Webhook integrations (Slack, Discord notifications)