Live Voice Chat AI

A real-time voice chat web app with streaming audio, live transcription, and low-latency responses. live url: https://live-voice-chat-ai.onrender.com

Highlights

• Full-duplex audio streaming (talk and listen without turn-taking)
• Live transcript feed
• Modern single-page UI (no frontend build step)

Stack

• Frontend: React (CDN), vanilla CSS, Opus recorder/decoder
• Backend: FastAPI for static assets + websocket server for streaming
• Model: Kyutai Labs Moshi (speech-to-speech)

Project Structure

• src/frontend/ UI (served as static files)
• src/app.py FastAPI server for static assets
• src/moshi.py websocket server for model streaming

Quick Start (Modal)

1. Install Modal

   pip install modal

2. Authenticate

   modal setup

3. Run the app

   modal serve src.app

Render (Frontend Only)

You can host the UI on Render and keep the websocket backend on Modal.

1. Deploy the backend on Modal

   modal deploy src.app

2. Create a Render Static Site connected to this repo
3. Set the Render publish directory to src/frontend
4. Set the websocket endpoint in src/frontend/index.html via window.LIVE_VOICE_WS_ENDPOINT, or pass it at runtime like https://your-render-site?ws=wss://YOUR-MODAL-ENDPOINT/ws.

Notes

• Microphone access is required in the browser.
• If you update frontend assets and don’t see changes, clear your browser cache.

Credits

Developed by Anurag Singh