Local TTS (Piper)

Fast local text-to-speech using Piper ONNX voices. Runs entirely offline with no cloud dependencies. Supports multiple languages and voice styles.

Usage

# Default voice (en_US-amy-medium)
~/.openclaw/skills/piper-tts/scripts/piper-tts.py "Hello, how are you today?"

# Select a specific voice
~/.openclaw/skills/piper-tts/scripts/piper-tts.py "Guten Tag" -v de_DE-thorsten-medium

# Pipe text from stdin
echo "Read this aloud" | ~/.openclaw/skills/piper-tts/scripts/piper-tts.py -

# Custom output path and format
~/.openclaw/skills/piper-tts/scripts/piper-tts.py "Hello" -o greeting.mp3 -f mp3

# Adjust speaking rate and send to Matrix room
~/.openclaw/skills/piper-tts/scripts/piper-tts.py "Slow and steady" --rate 0.8 --room-id '!abc:matrix.org'

# List available downloaded voices
~/.openclaw/skills/piper-tts/scripts/piper-tts.py --list-voices

# Quiet mode (suppress progress)
~/.openclaw/skills/piper-tts/scripts/piper-tts.py "Hello" --quiet

Options

• -v/--voice: Voice model name (default: en_US-amy-medium)
• -o/--output: Output file path (default: auto-generated in /tmp)
• -f/--format: Output format: wav, mp3, ogg (default: wav)
• --rate: Speaking rate multiplier, 0.5-2.0 (default: 1.0)
• --room-id: Matrix room ID to send audio to
• --list-voices: List downloaded voice models
• -q/--quiet: Suppress progress messages

Voices

Piper supports 900+ voices across 60+ languages. Voice models are auto-downloaded from HuggingFace on first use.

Popular voices

Benchmark

openclaw.json

{
  "tools": {
    "media": {
      "tts": {
        "enabled": true,
        "models": [
          {
            "type": "cli",
            "command": "~/.openclaw/skills/piper-tts/scripts/piper-tts.py",
            "args": ["--quiet", "-f", "ogg", "-o", "{{OutputPath}}", "{{Text}}"],
            "timeoutSeconds": 30
          }
        ]
      }
    }
  }
}