Live Translation System

Real-time sermon translation system that converts English speech to French audio in real-time.

Features

• Speech-to-Text: Deepgram Nova-2 with sermon-optimized keywords
• Translation: DeepL API (500K chars/month free)
• Text-to-Speech: Microsoft Edge TTS (100% FREE neural voices)
• Real-time: WebSocket-based streaming pipeline
• Low Latency: Optimized buffering for responsive translations

Architecture

Audio Input → Deepgram STT → DeepL Translation → Edge TTS → Audio Output
     ↓              ↓               ↓                ↓
  WebSocket    Nova-2 Model    EN → FR         Neural Voice

Quick Start

Prerequisites

• Node.js 18+
• Deepgram API key (for STT)
• DeepL API key (free tier available)

Installation

# Install dependencies
npm install

# Copy environment template
cp .env.example .env

# Add your API keys to .env
# DEEPGRAM_API_KEY=your_key_here
# DEEPL_API_KEY=your_key_here

# Start development server
npm run dev

Frontend

cd frontend
npm install
npm run dev

Configuration

Edit .env to configure:

Cost Estimate

~$2.60 per 10-hour sermon

Project Structure

src/
├── core/           # Core utilities (logger, event bus)
├── providers/      # STT, Translation, TTS providers
│   ├── stt/        # Deepgram STT
│   ├── translation/# DeepL
│   └── tts/        # Edge TTS, Deepgram TTS
├── services/       # Pipeline orchestrator, session manager
├── server/         # HTTP & WebSocket servers
└── types/          # TypeScript types

frontend/           # React/Vite frontend

License

MIT