Voxbento is an open-source, real-time interpretation platform designed for live events. It offers a browser-first, zero-install experience for simultaneous interpreters, enabling them to monitor event video via Jitsi and broadcast translated audio to attendees with sub-second latency. The platform benefits event organizers and interpreters by streamlining the interpretation workflow.

How It Works

The architecture prioritizes low-latency audio delivery using WebRTC. Interpreters connect via their browser, using Jitsi Meet to monitor the main conference feed. Their translated audio is captured and streamed using WHIP (WebRTC-HTTP Push) to MediaMTX, a dedicated WebRTC media server. MediaMTX handles the WebRTC termination and remuxing, then broadcasts the audio to attendees via WHEP (WebRTC-HTTP Playback), achieving sub-second latency. Concurrently, a FastAPI backend manages booth coordination, interpreter status, relay handoffs, and chat communication over WebSockets. This backend also orchestrates background transcription (via ffmpeg and services like Deepgram, OpenAI, or local models) and translation (using Groq, Anthropic, or Gemini APIs), crucially ensuring that Python code remains outside the direct audio processing path.

Quick Start & Requirements

• Primary Install: Docker Compose is the recommended setup.

  git clone https://github.com/fossasia/voxbento.git
  cd voxbento
  cp .env.example .env
  # Configure environment variables in .env (ADMIN_PASSWORD, DOCKER_HOST_ADDRESS)
  docker compose up --build
  

• Prerequisites: Python 3.13+, uv, MediaMTX (included in Docker), Docker & Docker Compose. For Jitsi video, the machine's LAN IP is required.
• Access: The platform is accessible at http://localhost:8000.
• Documentation: Official documentation is available at docs.voxbento.com.

Highlighted Details

• Real-time Interpretation: A browser-first, zero-install platform enabling simultaneous interpreters to participate remotely.
• Low-Latency Audio: Achieves sub-second latency for translated audio delivery to attendees using WebRTC (WHIP/WHEP) and the MediaMTX server.
• Integrated AI Services: Supports background transcription and translation via multiple third-party APIs (Deepgram, OpenAI, Groq, Anthropic, Gemini) and local models.
• Optional Hardware Acceleration: Includes optional support for NVIDIA Riva transcription models, allowing for reduced default installation footprint.

Maintenance & Community

Information regarding notable contributors, sponsorships, partnerships, community channels (Discord/Slack), or roadmaps is not detailed in the provided README.

Licensing & Compatibility

The license type is not explicitly stated in the provided README. Compatibility for commercial use or closed-source linking cannot be determined without explicit license information.

Limitations & Caveats

The platform requires specific environment variable configuration (ADMIN_PASSWORD, DOCKER_HOST_ADDRESS) for setup. Optional NVIDIA Riva support necessitates explicit installation, suggesting a larger default dependency if not managed. API key encryption requires generating and managing a secure API_KEY_ENCRYPTION_KEY.