Okeysir198 STT LiveKit Plugin
Build self-hosted Speech-to-Text systems for LiveKit voice agents using Whisper models from Hugging Face.
What's Included
This skill provides a complete self-hosted STT solution:
- STT API Service - FastAPI server running faster-whisper for efficient transcription
- LiveKit Plugin - Native LiveKit agents plugin for seamless integration
- Examples - Working examples of voice agents and transcription
- Documentation - Comprehensive guides and API documentation
Features
- 🚀 Fast transcription with faster-whisper (CTranslate2 optimization)
- 🔒 Self-hosted - full control over your data and infrastructure
- 🔄 Real-time streaming via WebSocket
- 📦 Batch processing via REST API
- 🌍 99+ languages with auto-detection
- 🐳 Docker-ready for easy deployment
Quick Start
- Start the STT API:
cd stt-livekit-plugin
docker-compose up -d
- Install the plugin:
cd livekit-plugin-custom-stt
pip install -e .
- Use in your voice agent:
from livekit.plugins import custom_stt
stt = custom_stt.STT(api_url="http://localhost:8000")
assistant = agents.VoiceAssistant(stt=stt, llm=..., tts=...)
Project Structure
stt-livekit-plugin/
├── stt-api/ # Self-hosted STT API service
│ ├── main.py # FastAPI application
│ ├── Dockerfile # Container image
│ └── requirements.txt
├── livekit-plugin-custom-stt/ # LiveKit plugin
│ ├── livekit/plugins/custom_stt/
│ │ ├── __init__.py
│ │ ├── stt.py # Main plugin implementation
│ │ └── version.py
│ ├── examples/ # Usage examples
│ └── pyproject.toml
├── docker-compose.yml # Easy deployment
└── README.md # Full documentation
Use Cases
- Voice assistants - Real-time conversation with AI agents
- Meeting transcription - Record and transcribe meetings
- Call centers - Analyze customer conversations
- Podcasts - Generate transcripts automatically
- Accessibility - Live captions and subtitles
Configuration
Choose the right model for your needs:
| Model | Speed | Accuracy | Best For |
|---|---|---|---|
| tiny | Fastest | Good | Real-time, low latency |
| base | Fast | Better | General purpose |
| small | Medium | Great | Balanced |
| medium | Slow | Excellent | High accuracy |
| large-v3 | Slowest | Best | Maximum accuracy |
Learn More
- Getting Started Guide
- STT API Documentation
- Plugin Documentation
- LiveKit Agents
Requirements
- Python 3.9+
- Docker (recommended)
- LiveKit server (for voice agents)
- GPU (optional, for better performance)