claude-speak

claude-speak

A Claude Code plugin that adds text-to-speech to Claude's responses using Kokoro TTS.

When enabled, Claude will speak notifications and response summaries aloud via a persistent local TTS server.

Prerequisites

• macOS (uses osascript for notifications, sounddevice for audio playback)
• uv
• espeak-ng — brew install espeak
• ffmpeg — brew install ffmpeg
• Kokoro TTS model files in ~/.local/share/kokoro-tts/:
  • kokoro-v1.0.onnx
  • voices-v1.0.bin

Installation

1. Install the Kokoro TTS tool:

   uv tool install kokoro-tts
   

2. Download model files to ~/.local/share/kokoro-tts/:

   mkdir -p ~/.local/share/kokoro-tts
   # Download kokoro-v1.0.onnx and voices-v1.0.bin from:
   # https://github.com/hexgrad/kokoro/releases
   

3. Add the plugin to your Claude Code project or global settings:

   claude plugins add /path/to/claude-speak
   

Configuration

Environment variables:

Variable	Default	Description
CLAUDE_TTS_VOICE	am_eric	Kokoro voice ID
CLAUDE_TTS_SPEED	1.1	Speech speed multiplier

Usage

/speak command

Toggle TTS on or off during a session:

/speak          # Toggle on/off
/speak on       # Enable TTS
/speak off      # Disable TTS
/speak status   # Show current state

How it works

The plugin registers hooks for three Claude Code events:

• Notification — Speaks notification messages
• Stop — Speaks a brief summary of Claude's response (first ~200 chars, stripped of markdown)
• SessionEnd — Cleans up the TTS server process

On first speech request, the hook starts a persistent Kokoro TTS server that keeps the model loaded in memory for low-latency synthesis. The server automatically shuts down after 10 minutes of inactivity.

File locations

Path	Purpose
~/.cache/claude-speak/kokoro.sock	Unix socket for TTS server
~/.cache/claude-speak/kokoro-server.log	Server log file
~/.config/claude-speak/disabled	Sentinel file (exists = TTS disabled)