
BridgeTalk is a real-time multilingual voice conversation bridge designed to help people communicate naturally across language barriers. It enables two speakers to have a conversation in different languages, starting with Hindi ↔ English support. BridgeTalk uses AssemblyAI Realtime Speech-to-Text to stream spoken audio and produce live transcripts as the conversation happens. When a speaker finishes a turn, the finalized transcript is translated into the other speaker's language while preserving important conversational details such as names, dates, times, locations, and amounts. The translated response is displayed in the conversation interface and can optionally be spoken aloud using Text-to-Speech, allowing the other person to hear the translation instead of relying only on written text. The application also provides structured conversation insights that identify useful factual information from the conversation. A Demo Mode allows the complete conversation experience to be demonstrated without consuming live AssemblyAI credits. The architecture uses a React and Vite frontend with a Python FastAPI backend. Audio is streamed through WebSockets to AssemblyAI's Realtime API, while translation and conversation processing run through the backend. API keys remain securely isolated on the backend and are never exposed to the frontend. Bridge Talk demonstrates how real-time speech recognition, translation, and voice technologies can work together to create a more natural multilingual conversation experience.
30 Sep 2026