Voice Environment - Nexus Knowledge Base

Environment: Voice

Location: /opt/mcp-servers/voice/mcp_voice_server.py Version: 5.2.0 Status: ✅ WORKING

Purpose

Multi-provider TTS (Text-to-Speech) system with automatic fallback chain: - Primary: Inworld AI ($10/1M chars) - Fallback: ElevenLabs ($100/month) - Local fallback: Piper TTS (no internet required)

Tools (1 total)

Tool	Parameters	Description
voice	paragraphs (req), voice, force_piper	Speak to user via TTS

Parameters

paragraphs: Array of text strings (max 500 chars each)
voice: "default" (Lena), "lars" (Edward), or ElevenLabs voice ID
force_piper: Boolean to force local TTS (testing)

Voice Options

Voice	Provider	Character
default	Inworld	Lena v3 (female)
lars	Inworld	Edward (male)

Features

Parallel generation: All paragraphs generated simultaneously
Sequential playback: Audio queued in order via WebSocket
Auto-notes: Every voice call creates context.notes entry
Phonetic conversion: Numbers→words for better TTS

Architecture

voice.voice(paragraphs) 
    → Inworld/ElevenLabs API (or Piper fallback)
    → WebSocket server (localhost:8765)
    → Browser audio playback
    → context.notes entry saved

Output Format

{
  "success": true,
  "provider": "inworld",
  "voice": "default",
  "paragraphs_spoken": 1,
  "total_chars": 26,
  "fallback_used": false
}

Usage Example

gateway.run([{
    server: 'voice',
    tool: 'voice',
    args: {
        paragraphs: ['Hello Chris, I have completed the task.'],
        voice: 'default'
    }
}])

Fallback Chain

Try Inworld AI (primary)
If fails → ElevenLabs
If fails → Piper local TTS

Security Assessment

✅ API keys stored in credentials (locker) ✅ WebSocket on localhost only ✅ Text sanitized before TTS

Audited by Maverick (a_7yma) | Documented by Rocky (o_cq0c) | 2026-01-06