feat: mvp

README.md

@@ -0,0 +1,67 @@
+# Fucking Chinese Dramas dubbing pipeline
+
+dubs fucking chinese drama movies to Russian/English using AI.
+
+can translate 2-hour movie in ~10 mins
+
+## requirements
+
+- uv
+- ffmpeg (for audio extraction and video muxing)
+- macOS with Apple Silicon (MPS) or Linux/Windows (CPU) (not tested)
+
+## installation
+
+```bash
+uv sync
+```
+
+all models (SenseVoice, Silero) download automatically on first run.
+
+## configuration
+
+create a `.env` file in the project root:
+
+```
+GEMINI_API_KEY=your_key_here
+```
+
+## usage
+
+```bash
+. .venv/bin/activate
+python main.py
+```
+
+cli will guide you through:
+1. selecting a project (or creating a new one)
+2. choosing translation mode
+3. running the pipeline
+
+## project structure
+
+```
+projects/
+  your_project/
+    source.mp4           ← put your video here
+    ...                  steps caching
+    result_language.mkv  final video with dubbed audio
+```
+
+## model cache
+
+models are downloaded automatically on first run and stored in:
+
+| Model | Path | Size |
+|-------|------|------|
+| Silero TTS | `~/.cache/torch/hub/snakers4_silero-models_master/` | ~40 MB |
+| SenseVoice ASR | `~/.cache/modelscope/hub/models/iic/SenseVoiceSmall/` | ~900 MB |
+| FSMN-VAD | `~/.cache/modelscope/hub/models/iic/speech_fsmn_vad_zh-cn-16k-common-pytorch/` | ~4 MB |
+
+to clear cache:
+```bash
+rm -rf ~/.cache/torch/hub/snakers4_silero-models_master/
+rm -rf ~/.cache/modelscope/hub/models/iic/
+```
+
+✅ Certified neuroslop