# ===================================================================== # EdgeSanskrit — Requirements # ===================================================================== # Python 3.10+ required. # Install with: pip install -r requirements.txt # ===================================================================== # ── Phase 1: Kokoro CPU TTS ────────────────────────────────────────── # Kokoro (StyleTTS2-based, 82M params) for fast local CPU inference kokoro>=0.9.4 huggingface_hub torch numpy>=1.24.0 soundfile>=0.12.1 python-dotenv>=1.0.0 loguru # Phase 3 dependencies (Web UI) fastapi>=0.111.0 uvicorn>=0.30.1 python-multipart>=0.0.9 # ── Phase 2: IndicF5 Authentic Sanskrit Chanting ───────────────────── # IMPORTANT: You MUST use AI4Bharat's IndicF5 fork from GitHub. # The PyPI package `f5-tts` is different code and its infer_process CRASHES. # Install the exact production commit used by Vagdhenu: git+https://github.com/ai4bharat/IndicF5.git@13f7c4d627cc10111aea8fe9c0039462cacacdc7 # Vocos vocoder (mel-to-audio decoder, required by IndicF5) vocos>=0.1.0 # IndicF5 transformer backbone dependency x-transformers>=2.19.7 # Indic script transliteration (Sanskrit ↔ Kannada ↔ SLP1 etc.) # Required by vagdhenu/src/prep_text.py for Devanagari → Kannada routing indic-transliteration # Audio loading / processing librosa torchaudio # SafeTensors format for loading IndicF5 base model weights safetensors requests>=2.31.0