text-to-speech

name	Text-to-Speech
description	This skill should be used when speaking text aloud, managing TTS voices, or controlling audio playback. Triggers on requests to speak, read aloud, announce, use voice output, or manage TTS settings. <example> Context: User wants audio feedback user: "Say hello" </example> <example> Context: User wants to change voice user: "Switch to a different voice" </example> <example> Context: User wants to stop audio user: "Stop talking" </example> <example> Context: User wants to hear available voices user: "What voices do you have?" </example>
version	1.0.0

Tools Reference

MCP Tools (marauder server)

Tool	Purpose
mcp__plugin_marauder_core__speak	Speak text aloud (async playback)
mcp__plugin_marauder_core__stop	Stop currently playing audio
mcp__plugin_marauder_core__voices	List installed voice models
mcp__plugin_marauder_core__current	Show active voice and status
mcp__plugin_marauder_core__download	Download a piper voice from HuggingFace
mcp__plugin_marauder_core__test	Test a voice with sample text (sync)

---

Text-to-Speech

Voice output using Piper TTS via the marauder MCP server.

Architecture

• Engine: Piper TTS (local, fast neural TTS)
• Transport: MCP stdio server (marauder mcp --mode local)
• Voices: ONNX models stored locally
• Playback: Async via system audio

Quick Start

Speak Text

mcp__plugin_marauder_core__speak(text: "Hello, I'm ready to help.")

Stop Playback

mcp__plugin_marauder_core__stop()

Check Current Voice

mcp__plugin_marauder_core__current()

Voice Management

List Installed Voices

mcp__plugin_marauder_core__voices()

Returns installed voice models with:

• Name (e.g., en_US-lessac-medium)
• Language
• Quality level

Download New Voice

mcp__plugin_marauder_core__download(voice: "en_US-lessac-medium")

Downloads from HuggingFace's Piper voice repository.

Test a Voice

mcp__plugin_marauder_core__test(voice: "en_US-lessac-medium")

Speaks sample text and waits for completion (synchronous).

Voice Naming Convention

Piper voices follow the pattern:

{language}_{region}-{name}-{quality}

Examples:

Voice	Description
en_US-lessac-medium	US English, Lessac, medium quality
en_GB-alan-medium	British English, Alan voice
en_US-amy-low	US English, Amy, low quality (faster)
en_US-ryan-high	US English, Ryan, high quality

Quality levels:

• low - Fastest, smaller model
• medium - Balanced (recommended)
• high - Best quality, larger model

Configuration

The active voice is set via environment variable:

PERSONALITY_VOICE=bt7274

Or in .mcp.json:

{
  "mcpServers": {
    "local": {
      "command": "marauder mcp",
      "args": ["--mode", "local"],
      "env": {
        "PERSONALITY_VOICE": "bt7274"
      }
    }
  }
}

Playback Behavior

Async by Default

speak() starts playback and returns immediately. Audio plays in background.

Sync Testing

test() blocks until audio completes. Use for voice previews.

Stopping

stop() immediately halts any playing audio.

Queue Behavior

New speak() calls while audio is playing will queue. Use stop() first if you need to interrupt.

Best Practices

1. Keep text concise - Long text takes time to synthesize
2. Use stop() before new speech - Avoid overlapping audio
3. Test voices before switching - Ensure quality meets needs
4. Match voice to persona - Different voices for different contexts
5. Consider latency - First synthesis may be slower (model loading)

Common Patterns

Notification on Task Complete

# After completing a long task
mcp__plugin_marauder_core__speak(text: "Build completed successfully")

Voice Selection Flow

# List available voices
voices = mcp__plugin_marauder_core__voices()

# Test one
mcp__plugin_marauder_core__test(voice: "en_US-ryan-high")

# If satisfied, update config to use it

Interrupt and Speak

mcp__plugin_marauder_core__stop()  # Stop any current speech
mcp__plugin_marauder_core__speak(text: "Urgent: deployment failed")

Troubleshooting

No Audio

• Check system audio output
• Verify voice model is installed: voices()
• Check current voice status: current()

Slow First Synthesis

• Normal - model loads on first use
• Subsequent calls are faster

Voice Not Found

• Download with download(voice: "voice-name")
• Check exact spelling from HuggingFace

Related

• Persona system uses TTS for voice output
• Session hooks can trigger speech on events
• Memory stores voice preferences