ronniiii-i/voice-call-frontend — reverse-engineered prompt

Build me a clean React TypeScript frontend for LinguaCall, a simple web app for two people to have a live translated voice call.

The user should be able to enter their name, choose their language, optionally set the server WebSocket URL, then create a room or join one with a room code. There should be a waiting screen that shows the code to share, then a call screen when the other person joins.

During the call, capture the microphone, send audio to the backend over WebSocket, play translated audio coming back, and show both original and translated captions in a live transcript. Add mute, end call, peer left handling, and a way to download the transcript as a txt file. No accounts or downloads needed.

Use React, TypeScript, Vite, React Router, and CSS Modules. Keep the design polished but simple, mobile friendly, and don’t use a utility CSS framework. Follow the WebSocket protocol and audio behavior from the project README, and look up current docs online if needed.