Back to LuxASR
University of Luxembourg LuxASR Logo

LuxASR Live

Real-Time Speech Recognition for Luxembourgish

Production Dual View

LuxASR Live provides real-time speech-to-text transcription. Click Start Recording to stream audio from your microphone and receive transcriptions as you speak. Press Space to toggle recording or Esc to stop.

Connection: Disconnected
Recording: 00:00
Segments: 0
Language: LB

Live Transcription

Processing...

Settings

Translates each segment in real-time
Sends previous text to improve continuity
Length: 0 words
Resets: 0
Click "Start Recording" to begin
-
Avg Latency
-
RTF
WebSocket Endpoint
-

How It Works

LuxASR Live uses WebSocket streaming to send audio from your browser to our speech recognition server in real-time. The audio is processed using Voice Activity Detection (VAD) to intelligently segment speech, then transcribed using our state-of-the-art Luxembourgish ASR model.

Features

  • Real-time transcription – See results as you speak
  • Voice Activity Detection – Intelligent audio segmentation
  • Multiple languages – Luxembourgish, English, French, German, Spanish, Portuguese
  • Low latency – Optimized for responsive transcription
  • Download support – Save your transcriptions as text files

Technical Details

  • Audio format: 16-bit PCM, 16kHz mono
  • Chunk duration: 1-30 seconds (adaptive based on speech activity)
  • Silence detection: Automatically sends audio after 1.5 seconds of silence
  • Protocol: WebSocket with binary audio and JSON control messages
  • Backend: FastAPI + Uvicorn (async I/O optimized)

WebSocket API

Connect to the WebSocket endpoint for programmatic access:

wss://luxasr.uni.lu/ws/transcribe

Client → Server:

  • Binary frames: 16-bit PCM audio samples
  • JSON: {"type": "config", "language": "lb"} – Set language
  • JSON: {"type": "stop"} – End session and process remaining audio
  • JSON: {"type": "clear"} – Clear accumulated transcription

Server → Client:

  • {"type": "connected", "session_id": "...", "config": {...}}
  • {"type": "transcription", "text": "...", "accumulated_text": "...", "segment": 1, "metrics": {...}}
  • {"type": "error", "error": "..."}

REST Endpoints

  • GET /live/health – Health check
  • GET /live/config – Current configuration
  • GET /live/sessions – Active sessions

Privacy & Data

Audio is streamed directly to our server at the University of Luxembourg and processed in real-time. No audio or transcription data is stored after the session ends. All processing stays within the University's network infrastructure in Luxembourg.

Browser Requirements

LuxASR Live requires a modern browser with WebSocket and Web Audio API support. Recommended browsers: Chrome, Firefox, Safari, Edge (latest versions).