fossasia/voxbento

What it solves

Voxbento is a real-time interpretation platform designed for live events. It eliminates the need for specialized hardware installations by providing a browser-based experience where simultaneous interpreters can monitor event video and broadcast translated audio to attendees with very low latency.

How it works

The system uses a combination of WebRTC, WHIP, and WHEP protocols via MediaMTX to handle audio streaming with sub-second latency. Interpreters use a browser-based interface to monitor the conference floor via Jitsi and stream their translated audio. A FastAPI portal manages coordination, state, and authentication. For automated features, a "floor-bot" joins the Jitsi meeting to feed audio into ffmpeg for background transcription and translation using providers like Deepgram, OpenAI, Groq, Anthropic, or Gemini.

Who it’s for

Event organizers and simultaneous interpreters who need a scalable, zero-install solution for live translation and audio broadcasting.

Highlights

• Zero-install experience: Works entirely in the browser for both interpreters and attendees.
• Low-latency audio: Utilizes WebRTC/WHIP/WHEP for near-instantaneous audio delivery.
• Flexible synchronization: Allows organizers to add listener-side audio delays to align translated audio with delayed video streams.
• AI Integration: Supports background transcription and translation via multiple third-party AI providers or local NVIDIA Riva models.