voice-reply

Local Text-to-Speech for OpenClaw using Piper voices via sherpa-onnx

Generate voice audio replies that work as Telegram voice notes - 100% offline, no API keys required.

Features

• 100% Local - No internet connection required after setup
• No API Keys - Completely free, no accounts needed
• Multi-language - German and English voices included (more available)
• Telegram Ready - Outputs as voice bubbles, not file attachments
• Auto-detect Language - Automatically selects the right voice

Quick Start

1. Install Dependencies

cd scripts
sudo ./install.sh

This installs:

• sherpa-onnx runtime (~28 MB)
• German voice "thorsten" (~64 MB)
• English voice "ryan" (~110 MB)
• ffmpeg (if not present)

2. Add to OpenClaw

Copy the skill to your OpenClaw skills directory:

cp -r . ~/.openclaw/skills/voice-reply

3. Use It

Ask your OpenClaw agent:

• "Reply with a voice message"
• "Say that as audio"
• "Read this aloud: Hello world"

Or call directly:

/voice_reply "Hello, how are you?" en

Voices

More voices available at Piper Samples.

Requirements

• Linux (Ubuntu 22.04+ recommended)
• ~200 MB disk space
• ~500 MB RAM during synthesis
• ffmpeg

How It Works

1. Text is converted to speech using sherpa-onnx with Piper VITS models
2. WAV output is converted to OGG Opus (Telegram voice format)
3. Output includes [[audio_as_voice]] tag for Telegram voice bubbles

Credits

• sherpa-onnx - Offline speech processing
• Piper - Fast local neural TTS
• Thorsten Voice - German voice dataset (CC0)

License

MIT License