VoiceClaw provides fully local, offline speech-to-text and text-to-speech capabilities for Openclaw Skills using Whisper and Piper.
Install & Download
1. ClawHub CLI
The fastest way to install a skill directly from the registry.
npx clawhub@latest install voiceclaw
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 voiceclaw 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 VoiceClaw?
VoiceClaw is a specialized skill for Openclaw Skills that brings high-performance, on-device voice processing to AI agents. By leveraging whisper.cpp for transcription and Piper for speech synthesis, it allows developers to build agents that can hear and speak without ever sending data to the cloud. This ensures maximum privacy, zero latency from network calls, and no reliance on third-party API keys.
The skill is designed for seamless integration into agentic workflows, handling everything from raw audio conversion with ffmpeg to generating natural-sounding responses in various English accents. It is the ideal solution for developers building secure, offline-first voice assistants or productivity tools within the ecosystem of Openclaw Skills.
VoiceClaw Use Cases
- Processing inbound voice messages from chat platforms like Telegram or Signal
- Building hands-free AI assistants that respond via voice
- Transcribing audio attachments for agent analysis and summarization
- Generating high-quality speech for automated notifications or status updates
- Ensuring data privacy in voice-controlled applications by avoiding cloud STT/TTS services
How VoiceClaw Works
- The agent identifies an inbound audio file or a request to speak.
- For transcription, the agent passes the audio file to the transcribe.sh script, which uses a local Whisper binary to generate text.
- The agent processes the resulting transcript as standard text input.
- To respond, the agent generates a text string and passes it to the speak.sh script.
- The Piper engine synthesizes the text into a WAV file using local ONNX voice models.
- Ffmpeg encodes the audio into an optimized format (e.g., OGG Opus) for delivery to the end user.
VoiceClaw Setup
To use this skill with Openclaw Skills, ensure you have the required binaries installed on your system:
# Check for required binaries
which whisper piper ffmpeg
# Set model paths if not using defaults
export WHISPER_MODEL="/path/to/ggml-base.en.bin"
export VOICECLAW_VOICES_DIR="/path/to/voices/"
# Test a voice synthesis
bash scripts/speak.sh "Hello from Openclaw Skills" output.wav
VoiceClaw Data Schema & Taxonomy
VoiceClaw organizes its operations around local binary paths and model files. Data flow is typically structured as follows:
| Component | File Types | Metadata/Config |
|---|---|---|
| STT (Whisper) | ogg, mp3, m4a, wav, flac | WHISPER_BIN, WHISPER_MODEL |
| TTS (Piper) | wav, onnx | PIPER_BIN, VOICECLAW_VOICES_DIR |
| Conversion | libopus, libmp3lame | Managed via ffmpeg flags |
| Agent I/O | Plain text transcripts | Integrated into agent message history |
VoiceClaw Advanced Features
- Support for multiple voice profiles including neutral American, warm female, and various British accents
- Local-first architecture with no network calls, perfect for air-gapped or privacy-sensitive environments
- Automated agent rules that prioritize voice output whenever voice input is received
- Optimized conversion for messaging platforms, including OGG Opus formatting for Telegram compatibility
- Custom model support allowing users to swap between different Whisper model sizes for speed vs. accuracy
SKILL.md
Loading
Related Openclaw Skills
OpenCLAW Tour Planner
A universal travel planning skill that enables AI agents to generate detailed itineraries, weather forecasts, and budget estimates through natural conversation.
OpenClaw Skill Lazy Loader
A performance optimization tool that reduces token usage by loading agent skills and context files only when required by the current task.
Humanize AI Content
A specialized tool for transforming sterile AI-generated text into high-quality, natural-sounding, and publication-ready editorial content.
SeedDance AI Video Generation
A professional-grade integration for ByteDance's SeedDance API, enabling programmatic text-to-video and image-to-video generation.
WordPress Publisher
A sophisticated automation skill for publishing Markdown content directly to WordPress with full Gutenberg block support and intelligent SEO optimization.
Devvit Publishing Auditor
A specialized pre-flight auditor for Reddit Devvit developers to verify app readiness and publishing compliance.
Featured*
