SpoofLab: a framework for audio deepfake and spoofing detection evaluation

The repository provides implementations and evaluation pipelines for analyzing the robustness and fairness of anti-spoofing models under controlled experimental conditions. It includes support for multiple front-end representations (e.g., LFCC, eGeMAPS, and self-supervised embeddings), as well as integration with state-of-the-art architectures such as AASIST and Conformer-based models.

The framework was used to conduct experiments on datasets such as ASVspoof and MLAAD, enabling investigations into bias related to language, gender, and prosodic variability in synthetic speech detection. All data and code are publicly available at: github.com/AI-Unicamp/SpoofLab No data files are hosted directly in this repository; all resources are maintained externally in the GitHub repository above.