How to Run an AI Text-to-Speech Agent with OE Runtime

Convert written content into natural-sounding speech audio automatically. OE Runtime connects to ElevenLabs, prepares your text for audio delivery, generates the audio with your chosen voice settings, and returns the file URL — all from a single YAML file.

What You Need

• Node.js 18+ installed, or the OE Runtime binary from openenthrium.com/runtime
• An API key from OpenAI, Anthropic, or any supported LLM provider
• An ElevenLabs API key for text-to-speech generation
• A text editor — nothing else

Create the Project Folder

Create a folder called speech-audio/ and add these two files:

The Skill Files

Create SKILL.md inside a speech-audio/ directory. This is the portable skill — it follows the agentskills.io format and runs unchanged on Claude, Cursor, Windsurf, or OE Runtime:

---
name: speech-audio
description: Convert text to natural-sounding speech audio
license: Apache-2.0
metadata:
  author: Open Enthrium
  version: "1.0"
---

You are a speech synthesis agent. Convert provided text to audio using ElevenLabs.
Choose a clear, professional voice appropriate for the content type.
Complete all steps fully before writing your report.

## Step 1: Prepare Content
Prepare the following text for speech synthesis. Clean it for audio delivery —
expand abbreviations and confirm it reads well aloud:

"Welcome to Open Enthrium. Your AI automation platform is ready. You can now
build, deploy, and run intelligent agents that connect to any tool or service.
Let's get started."

## Step 2: Select Voice
GET /voices to retrieve the list of available ElevenLabs voices.
Select a professional, clear voice suitable for corporate narration.
Save the voice_id of the selected voice.

## Step 3: Generate Audio
POST /text-to-speech/<voice_id> with body:
{
  "text": "<prepared text from step 1>",
  "model_id": "eleven_multilingual_v2",
  "voice_settings": { "stability": 0.7, "similarity_boost": 0.8 }
}
The response is an audio file saved to a local temp path. Report that path.

## Step 4: Report
Summarize the result:
- Text converted (character count)
- Voice selected (name and voice_id)
- Voice settings used (stability, similarity_boost)
- Output audio file path

Create agent.yaml in the same directory to wire the skill to your connector:

name: Text to Speech Agent
description: Convert text to natural-sounding speech audio
connectors:
  - connection_name: ElevenLabs
    connection_type: elevenlabs
skills:
  - path: ./
    trigger_type: auto

The Config File

Create oe-config.json in the same directory:

{
  "llm": {
    "provider": "openai",
    "model": "gpt-4o",
    "apiKey": "YOUR_OPENAI_API_KEY"
  },
  "server": {
    "enabled": false,
    "port": 3333,
    "apiKey": "your-secret-api-key"
  },
  "connectors": [
    {
      "connection_name": "ElevenLabs",
      "connection_type": "elevenlabs",
      "apiKey": "YOUR_ELEVENLABS_API_KEY",
      "headerName": "xi-api-key",
      "baseUrl": "https://api.elevenlabs.io/v1"
    }
  ]
}

Get your ElevenLabs API key from elevenlabs.io/app/settings. The agent automatically fetches available voices via GET /voices and selects the best match for the content — no need to hardcode a voice ID. Browse the full voice library at elevenlabs.io/voice-library. The generated audio is saved to a local file and the path is returned in the report.

Download OE Runtime

Run the Agent

From the parent folder containing your skill directory:

1 npx recommended

npx -y @openenthrium/oe-runtime@latest ./speech-audio

2 Windows

oe-runtime-win.exe ./speech-audio

3 macOS

chmod +x oe-runtime-macos
./oe-runtime-macos ./speech-audio

First run blocked? System Settings → Privacy & Security → Allow Anyway.

4 Linux

chmod +x oe-runtime-linux
./oe-runtime-linux ./speech-audio

5 API Server integration

Add a "server" block to oe-config.json, then start with --serve:

{
  "llm": { ... },
  "server": { "enabled": true, "port": 3333, "apiKey": "your-secret-key" },
  "connectors": [ ... ]
}

npx -y @openenthrium/oe-runtime@latest --serve --config oe-config.json

Run with inline YAML:

curl -X POST http://localhost:3333/run \
  -H "Content-Type: application/json" \
  -H "X-API-Key: your-secret-key" \
  -d '{"yaml": "...", "params": {}}'

Or run from a file on the server:

curl -X POST http://localhost:3333/run-file \
  -H "Content-Type: application/json" \
  -H "X-API-Key: your-secret-key" \
  -d '{"file": "/path/to/agent.yaml", "params": {}}'

Use Cases

• Podcast and audio content production — Convert written articles, blog posts, or scripts into professional-quality audio narrations for podcasts, audio blogs, or newsletter audio editions. The agent handles text preparation, normalises punctuation for natural pacing, and delivers a finished audio file ready for publishing.
• E-learning and training narration — Generate voiceovers for online course videos, training modules, and instructional content without recording studio time or voice talent fees. The agent produces consistent audio quality across hundreds of modules and can be re-run instantly whenever course content is updated.
• Accessibility and inclusive design — Create audio versions of written content — product documentation, help articles, and announcements — to make your product accessible to users with visual impairments or reading difficulties. Audio content also improves accessibility for users in hands-free or mobile contexts.
• IVR and voice assistant scripts — Generate natural-sounding audio prompts for interactive voice response systems, call centre greetings, and voice assistant responses. The agent can produce dozens of variations of the same prompt to A/B test which voice style and phrasing gets the best caller response.
• Video and presentation voiceovers — Add professional narration to marketing videos, product demos, explainer animations, and investor presentations without hiring a voice actor. The agent converts your script to audio in minutes and you can regenerate specific sections instantly if the copy changes.
• Multilingual content delivery — Generate audio in multiple languages using ElevenLabs' multilingual model to serve international audiences with localised voice content. Combined with a translation step, the agent can produce the same content in 29 languages in a single pipeline run.
• Automated customer communications — Convert personalised written messages — appointment reminders, delivery notifications, or account alerts — into audio for delivery via phone or voice messaging. Voice messages have significantly higher engagement rates than text messages for time-sensitive communications.
• Real-time content narration — Run the agent as an API server and call it from your CMS or publishing platform to automatically generate audio versions of every new article or post. Readers who prefer audio can listen rather than read, and your content library builds an audio catalogue passively over time.