A high-performance, on-device text-to-speech engine that gives AI agents a natural-sounding voice with ultra-low latency.
Install & Download
1. ClawHub CLI
The fastest way to install a skill directly from the registry.
npx clawhub@latest install her-voice
2. Manual Installation
Copy the skill folder to one of these locations
~/.openclaw/skills/ <project>/skills/ Priority: Workspace > Local > Bundled
3. Prompt Installation
Copy this prompt to OpenClaw to install it automatically.
Help me install her-voice using Clawhub. If Clawhub is not installed, install it first (npm i -g clawhub).
Prefer to download?
Get the raw skill files in a ZIP archive.
What is Her Voice?
Her Voice is a sophisticated text-to-speech skill designed to provide Openclaw Skills with expressive, local audio capabilities. Built on the Kokoro TTS model, it allows agents to communicate verbally using a compact 82M parameter model that runs entirely on-device, ensuring 100% privacy and no reliance on external APIs.
Optimized for speed, the skill utilizes on-the-fly audio streaming and an optional TTS daemon to minimize response times. On macOS, it leverages Apple's MLX framework for Metal acceleration and includes a native real-time visualizer, while providing a robust PyTorch fallback for Linux and Intel-based systems. This makes it a premier choice for developers looking to add personality and accessibility to their agent workflows.
Her Voice Use Cases
- Providing verbal responses to user queries in real-time.
- Reading aloud long-form content, documentation, or notifications.
- Creating accessible interfaces for users who prefer auditory feedback.
- Developing interactive voice-based stations using the persistent visualizer mode.
- Customizing agent personality through unique voice blending and phonetic name pronunciation.
How Her Voice Works
- The agent triggers the skill via a CLI command, passing the text to be spoken.
- A notification sound is optionally played immediately to provide instant feedback while the model initializes.
- The Kokoro TTS model generates audio chunks on-the-fly using either the MLX or PyTorch engine.
- If the TTS daemon is running, the system connects via a Unix socket to bypass model loading overhead.
- The generated audio is streamed to the system speakers or a native macOS visualizer for real-time playback.
Her Voice Setup
To get started with this skill for Openclaw Skills, run the automated setup wizard to install dependencies like espeak-ng and the Kokoro model:
python3 SKILL_DIR/scripts/setup.py
After installation, configure your agent's identity and phonetic pronunciation to ensure natural interaction:
python3 SKILL_DIR/scripts/config.py set agent_name "Jackie"
python3 SKILL_DIR/scripts/config.py set user_name_tts "Mah-toosh"
Her Voice Data Schema & Taxonomy
The skill maintains a dedicated configuration directory at ~/.her-voice/ to manage state and assets:
| Item | Description |
|---|---|
config.json |
Stores voice settings, speed, names, and visualizer preferences. |
tts.sock |
Unix socket used by the daemon for fast inter-process communication. |
bin/ |
Contains the compiled native macOS visualizer binary. |
voices/ |
Stores pre-blended voice weights in .safetensors format. |
venv/ |
Isolated Python environments for MLX or PyTorch backends. |
Her Voice Advanced Features
- TTS Daemon: Keeps the model warm in RAM to reduce latency by approximately 1.1 seconds per call.
- Voice Blending: Create unique vocal identities by mixing multiple base voices with custom weight distributions.
- Persist Mode: Maintains the visualizer on-screen for drag-and-drop audio playback and clipboard-to-speech functionality.
- Native Metal Acceleration: Uses Apple Silicon's GPU via MLX for highly efficient on-device synthesis.
- Phonetic Overrides: Configure phonetic spellings to ensure the engine correctly pronounces non-English names.
SKILL.md
Loading
Related Openclaw Skills
Briefing Room
An automated news station that researches global events and produces a conversational 10-minute audio briefing and document every day.
URL to Markdown Converter
A technical tool for fetching web URLs and converting their HTML content into clean, readable Markdown formats optimized for AI agents.
SEO Audit
A powerful SEO analysis tool for Openclaw Skills that identifies technical gaps and content opportunities to improve search engine rankings.
Office Docs
A comprehensive toolkit for programmatically creating, reading, and manipulating professional office documents including PDF, Word, Excel, and PowerPoint.
Agent Swarm Orchestrator
A high-performance orchestration framework that automates the full coding lifecycle from Obsidian task intake to GitLab merge using AI agents.
Polymarket Autopilot (Experimental)
An experimental autonomous agent for analyzing Polymarket public data and simulating paper trading with strict LLM cost controls.
Featured*
