B-Translate

Real-time AI-powered simultaneous translation platform for live events and conferences.

What It Does

A speaker talks into a microphone. Participants join via QR code or event code on their phones, pick their language, and instantly see — and hear — the live translation in under 1 second.

Architecture (V2)

Host Microphone
     │  (raw PCM 16kHz mono)
     ▼
  WebSocket
     │
     ▼
Deepgram Nova-3  ──→  STT (streaming, <300ms latency, built-in VAD)
     │
     ▼
 GPT-4o-mini     ──→  Context-aware translation
     │
     ▼
 Deepgram Aura   ──→  TTS (studio-quality audio, <250ms first byte)
     │
     ▼
  WebSocket
     │
     ▼
Participant Screen + Audio

Falls back to OpenAI Whisper + browser TTS automatically if DEEPGRAM_API_KEY is not set.

How It Works

1. Host creates a session, gets a QR code, starts broadcasting.
2. Audio is captured as raw PCM (16kHz, mono) and streamed continuously to the server.
3. Deepgram Nova-3 transcribes speech in real-time with built-in VAD — no client-side silence detection needed, zero hallucination.
4. GPT-4o-mini translates each finalized transcript to every requested language (once per language, shared across listeners).
5. Deepgram Aura converts the translated text to studio-quality MP3 audio on the server.
6. Participants receive both text and audio via WebSocket. Audio plays automatically if enabled.

Tech Stack

Layer	Technology
Backend	Python, FastAPI, WebSockets
STT	Deepgram Nova-3 (streaming) / OpenAI Whisper (fallback)
Translation	OpenAI GPT-4o-mini
TTS	Deepgram Aura / Browser speechSynthesis (fallback)
Frontend	Vanilla HTML/CSS/JS
Auth	JWT + bcrypt
Hosting	Render.com (CI/CD via GitHub)

Supported Languages

English, Turkish, German, French, Spanish, Arabic, Urdu, Chinese, Japanese

Quick Start

# 1. Clone & install
git clone https://github.com/ardacetin/btranslate.git
cd btranslate
pip install -r backend/requirements.txt

# 2. Set environment variables
export OPENAI_API_KEY="sk-..."
export DEEPGRAM_API_KEY="..."
export JWT_SECRET_KEY="your-secret"

# 3. Run
cd backend
uvicorn main:app --host 0.0.0.0 --port 8000

Open http://localhost:8000 in your browser.

Environment Variables

Variable	Required	Description
OPENAI_API_KEY	Yes	OpenAI API key — powers GPT-4o-mini translation
DEEPGRAM_API_KEY	Recommended	Deepgram API key — powers Nova-3 STT + Aura TTS
JWT_SECRET_KEY	Yes	Secret for JWT token signing
ADMIN_PASSWORD	No	Initial admin password (defaults to admin123)

Get a free Deepgram key at console.deepgram.com — includes $200/month free credit.

Project Structure

btranslate/
├── backend/
│   ├── main.py          # FastAPI routes & WebSocket endpoints
│   ├── ai_engine.py     # Deepgram STT/TTS, Whisper fallback, GPT translation
│   ├── sockets.py       # Connection manager & broadcast pipeline
│   ├── auth.py          # JWT authentication
│   └── requirements.txt
└── frontend/
    ├── index.html        # Landing page (Join as participant)
    ├── host.html         # Host dashboard (QR, broadcast controls)
    ├── participant.html  # Live feed (text + audio)
    ├── styles.css        # Design system
    └── lang.js           # i18n (EN/TR)

License

This project is licensed under CC BY-NC-SA 4.0.

Beykoz University — IT Directorate