The Rosetta Engine

Decode the language of the pharaohs — hear it spoken aloud.

An AI-powered Ancient Egyptian hieroglyph explorer that lets you browse, translate, and hear the language of the Nile. Drag glyphs onto a cartouche board, type English to get hieroglyphs, discover lost words, and listen to reconstructed Egyptian pronunciation through neural voice synthesis.

---

Features

• Hieroglyph Explorer — Browse 700+ hieroglyphs from the Gardiner Sign List, filtered by category and type (uniliteral, biliteral, triliteral, determinative, ideogram)
• Cartouche Board — Drag and drop glyphs to compose words and phrases, reorder with drag, see live transliteration
• English-to-Hieroglyph Translation — Type English words or phrases and get hieroglyphic translations with AI assistance
• Voice Synthesis — Hear reconstructed Ancient Egyptian pronunciation via XTTS v2 neural text-to-speech with a Coptic-trained reference voice
• Lost Words Engine — AI-powered discovery of unattested Egyptian words by analyzing Afro-Asiatic language patterns
• Semitic Vocalization — Reconstructs missing vowels using comparative Semitic linguistics (Arabic, Hebrew, Akkadian cognates)
• Random Word & Sentence Discovery — Explore 200+ Egyptian words and 90+ sentence templates spanning spells, prayers, royal decrees, love poetry, and medical texts
• Phoneme Engine — Converts transliterated Egyptian into speakable pronunciation using linguistic reconstruction rules
• Dark Egyptian Theme — Gold-on-dark UI with hieroglyph fonts, glow effects, and cartouche styling

Tech Stack

Layer	Technology
Frontend	React 19, Vite 6, Tailwind CSS 3.4
Backend	Python 3.11+, FastAPI, Uvicorn
Voice	Coqui XTTS v2 (neural TTS, GPU accelerated)
AI	Ollama (local LLM for translation & lost words)
Data	Gardiner Sign List, TLA Corpus (5,030 lemmas)
Font	Noto Sans Egyptian Hieroglyphs (Google Fonts)

Screenshots

Screenshots coming soon — run the app to see it in action!

Quick Start

One-Click (Windows)

Double-click start.bat — launches both servers and opens the browser.

Manual Setup

Prerequisites: Python 3.11+, Node.js 18+, Ollama (optional, for AI features)

Backend:

cd backend
python -m venv venv
venv\Scripts\activate                   # Windows
# source venv/bin/activate              # macOS/Linux
pip install torch torchvision torchaudio --index-url https://download.pytorch.org/whl/cu128
pip install -r requirements.txt
python main.py                          # Starts on port 8042

XTTS v2 model loads in the background (~30s, first run downloads ~1.8GB).

Frontend:

cd frontend
npm install
npm run dev                             # Starts on port 5173

Open http://localhost:5173 in your browser.

Voice Reference

Place a .wav file in backend/audio/reference/ for voice cloning. XTTS v2 clones that voice for all generated speech. Coptic and deep male reference voices are included.

Architecture

┌─────────────────────────────────────────────────┐
│                   Browser                        │
│  React 19 + Tailwind + Noto Egyptian Hieroglyphs │
│  ┌───────────┐ ┌──────────┐ ┌────────────────┐  │
│  │SymbolGrid │ │Cartouche │ │  MeaningPanel   │  │
│  │           │→│  Board   │→│  + AudioPlayer  │  │
│  └───────────┘ └──────────┘ └────────────────┘  │
└───────────────────────┬─────────────────────────┘
                        │ REST API
┌───────────────────────┴─────────────────────────┐
│              FastAPI Backend (:8042)              │
│  ┌──────────┐ ┌──────────┐ ┌──────────────────┐ │
│  │ /glyphs  │ │/translate│ │     /speak       │ │
│  │ /words   │ │  + LLM   │ │   XTTS v2 TTS   │ │
│  └────┬─────┘ └────┬─────┘ └───────┬──────────┘ │
│       │            │               │             │
│  ┌────┴────┐  ┌────┴──────┐  ┌────┴───────────┐ │
│  │Gardiner │  │ Translator│  │  Phoneme Engine │ │
│  │  Signs  │  │ Lost Words│  │  Semitic Vocal. │ │
│  │TLA Data │  │  Ollama   │  │  Coptic Ref WAV │ │
│  └─────────┘  └───────────┘  └─────────────────┘ │
└──────────────────────────────────────────────────┘

How the AI Features Work

Lost Words Engine

Discovers plausible but unattested Egyptian words by:

1. Analyzing patterns in the 5,030-lemma TLA corpus
2. Applying Afro-Asiatic root morphology rules
3. Using a local LLM to validate semantic plausibility
4. Generating pronunciation via the phoneme engine

Semitic Vocalization

Ancient Egyptian was written without vowels. The vocalization engine reconstructs them by:

1. Comparing Egyptian roots against cognates in Arabic, Hebrew, and Akkadian
2. Applying known vowel-pattern rules from Coptic (the last stage of Egyptian)
3. Using stress patterns from comparative Afro-Asiatic linguistics

Voice Synthesis

Pronunciation is generated through:

1. The phoneme engine converts transliterated Egyptian into IPA-like phonemes
2. XTTS v2 neural TTS synthesizes speech using a Coptic-trained reference voice
3. Output is served as WAV audio for in-browser playback
4. Conditioning latents are cached for instant repeat generation

Credits

• Thesaurus Linguae Aegyptiae (TLA) — Berlin-Brandenburg Academy of Sciences, 5,030 lemma corpus
• Gardiner Sign List — Sir Alan Gardiner's classification of Egyptian hieroglyphs
• Coqui XTTS v2 — Neural text-to-speech model
• Noto Sans Egyptian Hieroglyphs — Google Fonts
• Ollama — Local LLM inference

License

Apache License 2.0 — see LICENSE for details.