Church Live Translation

Real-time audio translation distribution system. A translator speaks into a mic in the booth — their audio streams live to listener phones, with live word-by-word transcription powered by Deepgram.

Project Structure

church-translator/
├── server/
│   ├── server.js        # Node.js WebSocket + Express server
│   └── package.json
└── public/
    ├── booth.html       # Translator input page
    └── listener.html    # Congregation output page

Setup & Run

1. Install dependencies

cd server && npm install

2. Set your Deepgram API key (required for transcription; audio relay works without it)

export DEEPGRAM_API_KEY=your_key_here

3. Start the server

node server.js

4. Open pages

• Translator booth: http://localhost:3000/booth.html
• Listener phones: http://localhost:3000/listener.html

Listeners must be on the same network. Use the machine's LAN IP (e.g. http://192.168.1.x:3000/listener.html), not localhost.

---

Production / Internet Access

Deploy the server/ directory to a VPS (Railway, Render, DigitalOcean, etc.).

The HTML files automatically use wss:// (secure WebSocket) when served over HTTPS — no code changes needed.

Set the DEEPGRAM_API_KEY environment variable in your hosting platform's dashboard.

---

How It Works

1. Booth captures mic audio via MediaRecorder (WebM/Opus, 250ms chunks) and sends binary chunks over WebSocket.
2. Server fans out the binary audio to all connected listener clients and simultaneously forwards it to Deepgram for transcription.
3. Listeners receive binary audio chunks, decode them with the Web Audio API, and play them back with a jitter buffer to avoid gaps. Transcript text appears line by line as Deepgram returns results.

iOS Note

Safari on iOS requires a user gesture to start audio playback. Listeners on iPhone/iPad will see a "Tap to Start Audio" prompt on first connection.