A SUPERB-Style Benchmark of Self-Supervised Speech Models for Audio Deepfake Detection

Spoof-SUPERB attempts to replicate the success of the original SUPERB (Speech processing Universal PERformance Benchmark) for the security-critical domain of deepfake detection. While the original SUPERB is a category-defining project (Score 9+), this specific extension is currently a nascent research artifact. With 0 stars and 4 forks, it lacks the community momentum and leaderboard infrastructure required to create a 'network effect' moat. Its primary value is the systematic evaluation of 20 different SSL models (like Wav2Vec 2.0 and HuBERT) on spoofing tasks, providing a comparative baseline that didn't exist in a unified format. The threat from frontier labs is medium; while labs like OpenAI and Google are incentivized to build internal detection benchmarks for safety (e.g., for Voice Engine), they often rely on third-party academic benchmarks for external validation. The project's defensibility is low because it is essentially a wrapper around existing models and datasets (like ASVspoof). Its longevity depends entirely on whether it becomes the official 'security' track for the broader SUPERB consortium. Without that institutional backing, it risks being a one-off paper repository displaced by the next major ASVspoof challenge cycle.