NathnelTK/TTS-Amharic

# 🇪🇹 Amharic TTS - የአማርኛ ድምጽ ሰሪ **Building the first truly natural-sounding Amharic Text-to-Speech system** * ለመጀመሪያ ጊዜ የአማርኛ ጽሑፍን ወደ ተፈጥሯዊ የሀበሻ ድምጽ * --- ## 🎯 The Problem | ችግሩ Current Amharic TTS systems sound like **"a foreigner trying to speak Amharic"** - unnatural, robotic, and hard to understand. ## 💡 Our Solution **Letter-by-letter learning approach:** - Each Ge'ez character recorded with its natural sound - Multiple native speakers with different accents - Character-level model (no complex rules needed) - Built by Ethiopians, for Ethiopians *እያንዳንዱን የግዕዝ ፊደል በተፈጥሯዊ ድምጹ እንቀዳለን* ## 🚀 Quick Start ### 1. Setup ```bash python setup_letter_folders.py ``` ### 2. Record - Record each Amharic letter clearly (0.5-1.5 seconds) - Multiple speakers preferred - Save in: `data/letter_sounds/{family}/{letter}/speaker_1/` ### 3. Train ```bash python scripts/prepare_training_data.py python scripts/train_tts_simple.py ``` ## 📊 What We Need ### Minimum for Demo - **50-100 letters** recorded - **1-2 speakers** - **3-5 hours** of recording - **Result**: Working prototype ### For Production - **All 231 Ge'ez characters** - **5+ speakers** (different regions/accents) - **20-30 hours** of recording - **Result**: Professional-quality TTS ## 🤝 **CALL FOR COMMUNITY SUPPORT ** ready to contribute then check out the contributing.md file ### 🌍 **Why This Matters ** **For Ethiopia:** - First world-class Amharic TTS - Accessibility for visually impaired - Educational technology advancement - Cultural and linguistic preservation - Ethiopian tech innovation showcase ### Installation ```bash pip install -r requirements.txt python setup_letter_folders.py ``` ### Project Structure | የፕሮጀክት አወቃቀር ``` ├── data/letter_sounds/ # Recording folders for 231 letters ├── scripts/ # Processing and training scripts ├── manifests/ # Training data manifests ├── models/ # Trained models └── README.md # This file ``` --- ## 🎯 Roadma …