# GlobalTalk AI
> One shared conversation. Every participant speaks and hears their own language.

## Overview
GlobalTalk AI is an open-source, production-ready real-time multilingual AI communication platform.
It eliminates language barriers across speech, text, documents, and video meetings with sub-second latency.

## Core Rules & Architecture
- Canonical Source Principle: Human speech or text is the only semantic source of truth. Translation chains (e.g. Hindi -> English -> Japanese) are forbidden; each target language fans out directly from the original source segment.
- WebRTC & WebSocket streaming pipeline: Audio Stream -> VAD -> STT -> NMT -> TTS -> Listener Routing.

## Key Capabilities
- Real-Time Voice Translation: Low-latency speech-to-speech translation with speaker diarization.
- GlobalTalk Write: DeepL-grade AI writing assistant with style and tone adaptation.
- Document Translation: Formats DOCX, PPTX, XLSX, PDF, TXT, HTML preserving layout.
- Multilingual Video Meetings: Powered by LiveKit WebRTC with per-listener translated audio streams.
- Multilingual Chat: Real-time chat translation in video rooms.
- Subtitle & Transcript Exporter: SubRip (.srt), WebVTT (.vtt), JSON, and Plain Text.

## Technical Specifications
- Frontend: React 18, TypeScript (strict), Vite, Tailwind CSS, Zustand, TanStack Query.
- Backend: Python 3.11, FastAPI, Pydantic v2, SQLAlchemy 2, Alembic.
- License: MIT Open Source.
- Repository: https://github.com/Aman678317/TalkFlow