Fetching metadata from the HF Docker repository...
-
fix: shorten README short_description for HF Spaces
-
chore: exclude sample audio files from git
-
chore: exclude large checkpoint files from git, preserve locally
-
Add Hugging Face Spaces deployment: Dockerfile + entrypoint + README YAML
-
Ignore TypeScript build artefacts
-
Initial commit: Audio Deepfake Detector with 8 detectors trained on jay15k
- 8 detectors: 2 production HF (MelodyMachine + motheecreator) + LF/HF physics + 4 paper architectures (Nes2Net, SONAR, BiCrossMamba-ST, VoiceRadar) + HoliAntiSpoof Qwen-stub
- All paper-architecture weights trained in-repo on deepfake_audio_dataset_jay15k
- Per-architecture hyperparameter profiles (LR, epochs, weight-decay)
- Logistic-regression meta-classifier for the ensemble verdict (test acc 0.983)
- Light/dark theme toggle with CSS variables
- 4-page UI: Home / Demo / Benchmarks / About
- FastAPI backend, React+TypeScript+Tailwind frontend
- Bundled ffmpeg for WebM/Opus audio decode (browser mic recordings)
