jsmeet/livekit-ai-subtitles repository overview

⁠LiveKit AI Subtitles

Real-time speech translation agent built on LiveKit Agents⁠.
Transcribes Hindi speech via Deepgram STT and streams translated subtitles to subscribers using OpenAI GPT.

⁠How It Works

1. Agent joins a LiveKit room and listens to participant audio
2. Deepgram transcribes speech in real-time (Hindi)
3. Subscribers send a subscribe message with their desired language
4. OpenAI translates and streams subtitle chunks back over the data channel

⁠Quick Start

docker run --env-file .env jsmeet/livekit-ai-subtitles:latest

⁠Environment Variables

⁠Docker Compose

services:
  ai-translation-service:
    image: jsmeet/livekit-ai-subtitles:latest
    restart: unless-stopped
    env_file: .env
    extra_hosts:
      - "host.docker.internal:host-gateway"

⁠Supported Languages

en hi mr ta te bn gu kn ml pa ur
es fr de zh ja ko ar ru it pt

⁠Data Channel Protocol

⁠Subscribe

{ "type": "subscribe", "lang": "en" }

⁠Incoming subtitles

{ "type": "subtitle_chunk", "lang": "en", "speaker": "Alice", "text": "Hello", "ts": 1234567890 }

⁠Tags