- AudioConfig now includes rep_max_count and rep_audio_dir
- app/audio/generate.py uses config instead of hardcoded constants
- RepAnnouncer rewrote with pre-generated audio cache
- Supports Windows winsound, macOS afplay, Linux paplay/aplay
- Pin requirements back to mediapipe==0.10.21 with numpy<2
- ServerConfig, VideoConfig, ModelConfig, DeadBugConfig,
AudioConfig, LoggingConfig as nested dataclasses
- Consumers use config.server.host, config.model.resolved_path etc.
- env var overrides preserved via _apply_env_overrides()
- All settings moved to config.yaml
- configs/load.py reads from config.yaml with env var overrides
- Environment variables still work for backward compatibility
- Added pyyaml to requirements
wsy182